Estimating Occupational Exposures with a Multi-Hazard ...

ESTIMATING OCCUPATIONAL EXPOSURES WITH A MULTI-HAZARD

SENSOR NETWORK

By

Christopher Zuidema

A dissertation submitted to Johns Hopkins University in conformity with the

requirements for the degree of Doctor of Philosophy

Baltimore, Maryland

July 2018

© 2018 Christopher Zuidema

All Rights Reserved

ii

ABSTRACT

Problem Statement: Exposure assessment and monitoring of occupational hazards is

typically performed to assess regulatory compliance, and almost exclusively relies on

personal sampling or measurement. However, personal measurements, primarily

conducted by industrial hygienists, can be expensive and burdensome and often suffers

from a low number of samples. Motivated to overcome the limitations of personal exposure

measurement, this dissertation instead proposed and investigated estimating personal

exposure with a multi-hazard sensor network.

Methods: In the first of three related manuscripts, we conducted a laboratory evaluation of

a low-cost sensor strategy to reduce the measurement error of quantifying ozone and

nitrogen dioxide concentrations with electrochemical sensors. Typical sensors for these

gases are in actuality “oxidizing gas” sensors, detecting both ozone and nitrogen dioxide

without discrimination. In the second manuscript, we reported on the long-term

deployment of a multi-hazard sensor network designed for this project that included

sensors for particulate matter, carbon monoxide, oxidizing gases, and noise. We assessed

the space-time variability of the hazards captured by the sensor network, and the accuracy

and precision of the sensor network measurements. In the third manuscript we developed

a technique to derive personal exposure estimates from the sensor network, simulated

facility employees while collecting personal measurements with field reference

instruments, and compared the network-derived personal exposure estimates to the

personal measurements.

iii

Results: In our first study, we observed measurement error for ozone was two to three

times higher than for nitrogen dioxide and that ozone was progressively underestimated as

the ratio of nitrogen dioxide to ozone increased. In our second study, we demonstrated the

first long-term deployment of a sensor network in a manufacturing setting capable of

measuring multiple hazards with a high degree of space-time resolution. The accuracy of

network measurements differed among the four hazards of interest, with the median percent

bias with reference to direct-reading instruments equal to 41%, 7%, 36% and 1%, for

particulate matter, carbon monoxide, oxidizing gases and noise respectively. Network

sensors exhibited varying degrees of precision with 95% of measurements among 3

collocated nodes within 0.23 mg/m3 for particulate matter, 0.4 ppm for carbon monoxide,

7 ppb for oxidizing gases, and 1 dBA for noise of each other. In our third study, we

observed the difference and correlation between personal exposure measurements and

network-derived personal exposure estimates varied greatly between the hazard under

study. The best correlation was found for noise, with the Pearson correlation coefficient

equal to 0.75.

Conclusions: Low-cost sensors may be subject to high levels of measurement error,

principally related to sensitivity, responsiveness to non-target species, and signal

degradation over time. Ultimately, the success of our technique to estimate personal

exposures was highly dependent on the accuracy of the sensor network’s underlying

measurements. We have demonstrated that estimating personal exposure holds promise as

an additional tool to be used with traditional personal measurement due to the ability to

frequently and easily collect exposure data on many employees.

iv

COMMITTEE OF THESIS READERS &

FINAL ORAL EXAMINATION COMMITTEE

Advisor: Kirsten Koehler, PhD

Associate Professor of Environmental Health and Engineering

Readers: Ana Rule, PhD

Assistant Professor of Environmental Health and Engineering

Mary Fox, PhD, MPH

Assistant Professor of Health Policy and Management

Frank Curriero, PhD

Associate Professor of Epidemiology

Alternates: Peter Lees, PhD, CIH

Professor of Environmental Health and Engineering

Howard Katz, PhD

Professor of Materials Science and Engineering

RESEARCH COMMITTEE

Kirsten Koehler, PhD

Associate Professor of Environmental Health and Engineering

Ana Rule, PhD

Assistant Professor of Environmental Health and Engineering

Peter Lees, PhD, CIH

Professor of Environmental Health and Engineering

Thomas Peters, PhD, CIH

Professor of Occupational and Environmental Health at the University of Iowa

Geb Thomas, PhD

Professor of Mechanical and Industrial Engineering at the University of Iowa

v

ACKNOWLEDGEMENTS

I owe many people many thanks for my education and the research in this dissertation.

First, I must thank my advisor, Kirsten Koehler, for bringing me into her lab, teaching me,

and being so supportive over the last two years. I am especially thankful for her

unflappability and sense of what is important and what is not so important, especially when

things don’t go according to plan as they often do in the field. This project was a great

opportunity for a doctoral student, and I’m very appreciative to have had it. I hope we can

finish our ideas on this project together in the near future.

I’m sure I wouldn’t have had the opportunity to become a student at Hopkins without Peter

Lees. He took me on and has continued to mentor and advise me throughout the program.

Thank you!

Tom Peters and Geb Thomas, Professors at the University of Iowa, designed, built and

maintained our sensor network, provided field and logistical support, and contributed

thoughtful feedback on manuscripts. They worked on this project long before I arrived on

scene, and without them, I wouldn’t have had a project to work on. They have advised me

from a distance and ensured our fieldwork went smoothly. I’m very grateful to you both.

My Mom, Debbie, and Dad, Bill, have been there since (before) the beginning, or as they

like to say, BC (“before Chris”). They have offered me the best education and opportunities

parents could. Through a decade of college and graduate school they have loved and

vi

supported me and have never asked “when are you going to get a job?” Maybe now, we’ll

see.

Monika, my best friend and partner, has encouraged and supported me from the instant I

expressed a desire to return to school. Through the ups and downs, she has selflessly been

by my side. I remember her literally jumping up and down with excitement next to me

while I was on the phone with Peter when he called to tell me I had been admitted to the

program. She has helped me stay grounded, keep perspective about what is important, and

get outside to have fun. Here’s to the first time one of us hasn’t been in school for the last

ten years!

vii

FUNDING SUPPORT

This dissertation was funded under support from the Johns Hopkins University Education

and Research Center for Occupational Safety and Health (ERC). ERC training grant

funding comes from the National Institute for Occupational Safety and Health (NIOSH),

under Grant No. 5 T42 OH 008428. This project was also funded through NIOSH under

Grant No. R01 OH 010533.

viii

TABLE OF CONTENTS

ABSTRACT ....................................................................................................................... ii

COMMITTEE OF THESIS READERS & FINAL ORAL EXAMINATION

COMMITTEE .................................................................................................................. iv

RESEARCH COMMITTEE........................................................................................... iv

ACKNOWLEDGEMENTS ............................................................................................. v

FUNDING SUPPORT .................................................................................................... vii

TABLE OF CONTENTS .............................................................................................. viii

LIST OF TABLES ............................................................................................................ x

LIST OF FIGURES ......................................................................................................... xi

LIST OF APPENDICES ................................................................................................ xii

CHAPTER ONE: Introduction ....................................................................................... 1

RATIONALE FOR RESEARCH ................................................................................ 2

DISSERTATION AIMS & STRUCTURE ................................................................. 6

REFERENCES .............................................................................................................. 7

CHAPTER TWO: Efficacy of Paired Electrochemical Sensors for Measuring Ozone

Concentrations .................................................................................................................. 9

ABSTRACT ................................................................................................................. 10

INTRODUCTION....................................................................................................... 11

METHODS .................................................................................................................. 14

RESULTS .................................................................................................................... 23

DISCUSSION .............................................................................................................. 27

CONCLUSIONS ......................................................................................................... 32

TABLES & FIGURES ................................................................................................ 33

REFERENCES ............................................................................................................ 40

CHAPTER THREE: Mapping Occupational Hazards with a Multi-Sensor Network

in a Heavy-Vehicle Manufacturing Facility ................................................................. 43

ABSTRACT ................................................................................................................. 44

INTRODUCTION....................................................................................................... 45

METHODS .................................................................................................................. 49

RESULTS .................................................................................................................... 55

DISCUSSION .............................................................................................................. 60

CONCLUSIONS ......................................................................................................... 66

FIGURES ..................................................................................................................... 68

REFERENCES ............................................................................................................ 75

ix

CHAPTER FOUR: Estimating Personal Exposure with a Multi-Hazard Sensor

Network ............................................................................................................................ 75

ABSTRACT ................................................................................................................. 84

INTRODUCTION....................................................................................................... 85

METHODS .................................................................................................................. 88

RESULTS .................................................................................................................... 92

DISCUSSION .............................................................................................................. 96

TABLES & FIGURES .............................................................................................. 101

REFERENCES .......................................................................................................... 108

CHAPTER FIVE: Conclusion ..................................................................................... 108

SUMMARY FINDINGS ........................................................................................... 117

FUTURE RESEARCH, PUBLIC HEALTH IMPLICATIONS, AND

CONCLUDING REMARKS ................................................................................... 120

APPENDICES ............................................................................................................... 129

CURRICULUM VITAE ............................................................................................... 133

x

LIST OF TABLES

Table 2.1. Summary of least-squares linear regression ............................................... 33

Table 2.2. Sensor response during baseline .................................................................. 34

Table 2.3. Mean average percent error for NO2 and O3 ............................................. 35

Table 4.1. Low-cost sensors and reference direct-reading instruments .................. 101

Table 4.2. Comparison of reference direct-reading instrument measurements and

network-derived exposure estimates for the stationary routine ....................... 102

Table 4.3. Comparison of reference direct-reading instrument measurements and

network-derived exposure estimates for the mobile routine............................. 103

xi

LIST OF FIGURES Figure 2.1. Sensor setup.................................................................................................. 36

Figure 2.2. Setup used for the calibration experiments ............................................... 37

Figure 2.3. Bias and variation maps of NO2 and O3 concentration estimates for

Method 1 .................................................................................................................. 38

Figure 2.4. Bias and variation maps of NO2 and O3 concentration estimates for

Method 2 .................................................................................................................. 39

Figure 3.1. Time series of hazard concentrations/intensities measured by the multi-

hazard monitor network......................................................................................... 68

Figure 3.2. Distribution of hazard level by manufacturing process ........................... 69

Figure 3.3. Example hazard maps ................................................................................. 71

Figure 3.4. Precision of measurements among collocated monitors .......................... 72

Figure 3.5. Second-order coefficient of variation of three collocated sensors ........... 73

Figure 3.6. Sensor measurement accuracy ................................................................... 74

Figure 4.1. Schematic of technique to estimate personal exposure from sensor

network .................................................................................................................. 104

Figure 4.2. Examples of timeseries comparing network-derived exposure estimates

and reference DRI measurements ....................................................................... 105

Figure 4.3. Cummulative density function plots of the stationary routine .............. 106

Figure 4.4. Cummulative density function plots of the mobile routine.................... 107

xii

LIST OF APPENDICES Appendix 2.1. Experimental Data for Method 1. ....................................................... 130

Appendix 2.2. Experimental Data for Method 2 ........................................................ 131

Appendix 3.1. Pearson’s correlation between hazards and temperature. ............... 132

1

CHAPTER ONE

Introduction

2

RATIONALE FOR RESEARCH

Occupational environments, especially in heavy industry often have complex, hazardous

exposures resulting from welding and other metalworking processes. Exposures can vary

greatly depending on the type of welding (e.g. MIG versus TIG welding) and include

airborne chemical and particulate hazards such as carbon monoxide (CO), oxides of

nitrogen (NOx), ozone (O3), lead, nickel, zinc, iron oxide, copper, cadmium, fluorides,

manganese, chromium; physical hazards including noise, heat, electrical, vibration; and

radiological hazards like visible, ultraviolet and infrared frequencies of light (Sferlazza

and Beckett 1991). A variety of health effects are associated with welding including

adverse respiratory and neurological effects that range in severity from respiratory

irritation and infection and changes in pulmonary function to possibly lung cancer and

Parkinsonism (Antonini 2003). In addition to welding, common pollutant-generating

metalworking processes in heavy industry include torch and laser cutting, machining,

grinding, and abrasive blasting.

Employers are required to protect workers from hazardous exposures, such as those

related to metalworking, and maintain hazard levels below established regulatory

occupational exposure limits (OELS). In the United States, the applicable regulatory

OELs are the permissible exposure limits (PELS) established by the Occupational Safety

and Health Administration (OSHA). Commonly, other non-regulatory OELs such as the

threshold limit values (TLVs) established by the American Conference of Governmental

Industrial Hygienists (ACGIH) are used as “best-practice” OELs by employers. Exposure

monitoring and assessment are therefore generally performed by employers in order to

3

evaluate if they are in compliance with OELs. Industrial hygienists typically identify

similar exposure groups (SEGs) and target the highest exposed individuals for personal

sampling; although, the number of samples taken or employees sampled is not prescribed

by law (Rappaport and Kupper 2008). In personal sampling or measurement, an

employee wears equipment, such as a filter and sampling pump or a measurement device

such as a direct-reading instrument (DRI), throughout their work shift. In the case of

personal air sampling, for example, the inlet/sampling head is placed in the breathing

zone (Leidel 1977).

While personal sampling is the gold standard for assessing occupational exposure, it can

have drawbacks. Personal measurement is expensive and burdensome to employers and

employees and generally suffers from a low number of samples taken (Rappaport 1984).

In most cases, fewer than six samples at an industrial facility are used to judge if

employees may be over-exposed or facilities are in compliance with OELs (Roick et al.

1991), and many rely on just one measurement (Tornero‐Velez et al. 1997). Chiefly due

to exposure variability between- and within-workers (Kromhout et al. 1993; Rappaport et

al. 1993), as it is currently practiced, personal measurements may leave workplace

exposures inadequately characterized and risks higher than measurements indicate

(Rappaport 1984). Even from the introduction of the first personal sampling pumps that

ushered in the era of personal sampling, the prediction that personal samplers “…in their

present form…are probably unsuitable for routine assessment of the exposure of large

numbers of people” is an apt observation that is still relevant today (Sherwood and

Greenhalgh 1960).

4

In this dissertation, we conceived of and demonstrated an alternative to personal

measurement: estimating personal exposure with a distributed monitoring network. To

estimate personal exposure, we integrated two data streams: 1) space-time hazard data

from a multi-hazard sensor network, and 2) location information of the employee(s) for

which exposure is being estimated. In contrast to taking traditional personal

measurements, creating network-derived exposure estimates increases the practicality and

feasibility of collecting occupational exposure data on potentially any/all workers on a

daily basis.

For this project, we designed and built the first multi-hazard sensor network, comprised

of 40 individual “nodes,” constructed with sensors for particulate matter (PM), CO,

oxidizing gases (O3 + NO2) and noise (Thomas et al. 2018). The hazards in our sensor

network were identified by industrial hygiene assessments of the study site, a heavy-

vehicle manufacturing facility, and were chosen based on occupational health

importance. To be practical, we required all of the components for each node cost less

than $1000 and, as such, each of the sensors was ≤ $150. However, many low-cost

sensors suffer from lower sensitivity/specificity, exhibit cross-sensitivity with non-target

species, are subject to signal baseline drift over time and suffer from lower data quality

(Lewis and Edwards 2016; Lewis et al. 2016; Masson et al. 2015; Piedrahita et al. 2014;

Snyder et al. 2013). In fact, the oxidative gas sensor selected for our network to measure

ozone exhibited significant cross-interference with nitrogen dioxide. The first manuscript

of this dissertation assessed an industry solution for isolating the ozone measurement.

5

Despite potentially lower-quality data, utilizing space-time hazard data from a multi-

hazard sensor network has extended current methods in hazard mapping, which

characterizes and displays the spatial distribution of hazards throughout a facility or

geographic area (Koehler and Volckens 2011; Koehler and Peters 2013; O'Brien 2003;

Peters et al. 2006). Our network allowed us to measure multiple hazards simultaneously

across the study site and create hazard maps for each of the hazards for any time period

of interest during a continuous eight-month-long deployment. In the second manuscript

of this dissertation, we report on the long-term deployment of the sensor network at the

study site, characterized the space-time variability of hazards under study, and evaluated

the precision and accuracy of the sensor network’s measurements. In the third manuscript

of this dissertation, we report on the differences and correlation between the network-

derived exposure estimates and reference DRI exposure measurements.

In summary, the research in this dissertation contributes to the body of knowledge on

low-cost sensors, sensor networks, and occupational exposure assessment. This project

was innovative because we sought to provide rapid, low-cost estimates of personal

occupational exposure compared to current industrial hygiene methods. Compared to

personal measurements, network-derived exposure estimates have the advantage of being

non-specific to any particular individual and easily scalable to a large number of workers.

To take personal measurements, a worker must be equipped with the requisite equipment

and, depending on the number of agents of interest and the number of workers, this

quickly becomes logistically infeasible because workers could be wearing or carrying

6

numerous direct reading instruments or samplers and professional staff would need to set-

up, maintain and manage all of the equipment. In contrast, estimating personal exposures

is passive, utilizing stationary network nodes located throughout a facility and requiring

only estimates of worker locations. Future work will focus on the use of automated

indoor positioning systems to provide location information for deriving exposure

estimates with sensor networks. In this scenario, the number of positioning devices and

willing workers are the only limiting factors. Estimating personal exposure has the

potential to offer a wealth of information to examine hazards and reduce risk in

occupational settings.

DISSERTATION AIMS & STRUCTURE

The three aims of this dissertation were to:

1. Evaluate a low-cost sensor solution for quantifying NO2 and O3 concentrations in

mixture.

2. Establish sensor networks as useful tools for measuring occupational hazards with

a high degree of space-time resolution.

3. Develop a method to estimate personal exposure to occupational hazards and

compare traditional personal measurements to network-derived estimates.

This body of this dissertation is comprised of three related manuscripts (chapters 2-4)

each corresponding to one of the dissertation’s three specific aims. Chapter one serves as

introduction and chapter five as conclusion.

7

REFERENCES

Antonini JM. 2003. Health effects of welding. Critical reviews in toxicology 33:61-103.

Koehler KA, Volckens J. 2011. Prospects and pitfalls of occupational hazard mapping:

'Between these lines there be dragons'. The Annals of Occupational Hygiene 55:829-

840.

Koehler KA, Peters TM. 2013. Influence of analysis methods on interpretation of hazard

maps. The Annals of Occupational Hygiene 57:558-570.

Kromhout H, Symanski E, Rappaport SM. 1993. A comprehensive evaluation of within-

and between-worker components of occupational exposure to chemical agents. The

Annals of Occupational Hygiene 37:253-270.

Leidel NA. 1977. Occupational exposure sampling strategy manual.

Lewis A, Edwards P. 2016. Validate personal air-pollution sensors. Nature 535:29-31.

Lewis AC, Lee JD, Edwards PM, Shaw MD, Evans MJ, Moller SJ, et al. 2016.

Evaluating the performance of low cost chemical sensors for air pollution research.

Faraday discussions.

Masson N, Piedrahita R, Hannigan M. 2015. Quantification method for electrolytic

sensors in long-term monitoring of ambient air quality. Sensors 15:27283-27302.

O'Brien DM. 2003. Aerosol mapping of a facility with multiple cases of hypersensitivity

pneumonitis: Demonstration of mist reduction and a possible dose/response

relationship. Applied Occupational and Environmental Hygiene 18:947-952.

Peters TM, Heitbrink WA, Evans DE, Slavin TJ, Maynard AD. 2006. The mapping of

fine and ultrafine particle concentrations in an engine machining and assembly

facility. The Annals of Occupational Hygiene 50:249-257.

8

Piedrahita R, Xiang Y, Masson N, Ortega J, Collier A, Jiang Y, et al. 2014. The next

generation of low-cost personal air quality sensors for quantitative exposure

monitoring. Atmospheric Measurement Techniques 7:3325.

Rappaport SM. 1984. The rules of the game: An analysis of osha's enforcement strategy.

American Journal of Industrial Medicine 6:291-303.

Rappaport SM, Kromhouta H, Symanski E. 1993. Variation of exposure between workers

in homogeneous exposure groups. The American Industrial Hygiene Association

Journal 54:654-662.

Rappaport SM, Kupper LL. 2008. Quantitative exposure assessment:S. Rappaport.

Sferlazza SJ, Beckett WS. 1991. The respiratory health of welders1-3. Am Rev Respir

Dis 143:1134-1148.

Sherwood RJ, Greenhalgh DMS. 1960. A personal air sampler. Ann Occup Hyg 2:127-

132.

Snyder EG, Watkins TH, Solomon PA, Thoma ED, Williams RW, Hagler GSW, et al.

2013. The changing paradigm of air pollution monitoring. Environmental science &

technology 47:11369.

Thomas G, Sousan S, Tatum M, Liu X, Zuidema C, Fitzpatrick M, et al. 2018. Low-cost,

distributed environmental monitors for factory worker health. Sensors 18:1411.

Tornero‐Velez R, Symanski E, Kromhout H, Yu RC, Rappaport SM. 1997. Compliance

versus risk in assessing occupational exposures. Risk Analysis 17:279-292.

9

CHAPTER TWO

Manuscript 1

(In revision for publication in the Journal of Occupational and Environmental Hygiene)

Efficacy of Paired Electrochemical Sensors for Measuring

Ozone Concentrations

Christopher Zuidema, Nima Afshar-Mohajer, Marcus Tatum, Geb Thomas, Thomas

Peters, and Kirsten Koehler

10

ABSTRACT

Typical low-cost electrochemical sensors for ozone (O3) are also highly responsive to

nitrogen dioxide (NO2). Consequently, a single sensor’s response to O3 is

indistinguishable from its response to NO2. Recently, a method for quantifying O3

concentrations became commercially-available: collocating a pair of sensors, a typical

oxidative gas sensor that responds to both O3 and NO2 (model OX-B431, Alphasense

Ltd., Essex, UK) and a second similar sensor, equipped with a manganese dioxide filter

that removes O3 and responds only to NO2 (model NO2-B43F, Alphasense Ltd., Essex,

UK). By pairing the two sensors, ozone concentrations can be calculated. We calibrated a

sample of 3 NO2-B43F sensors and 3 OX-B431 sensors with NO2 and O3 exclusively and

conducted mixture experiments over a range of 0-1.0 ppm NO2 and 0-125 ppb O3 to

evaluate the ability of the paired electrochemical sensors to quantify NO2 and O3

concentrations in mixture. Although the slopes of the response for each sensor varied, the

individual response of the NO2-B43F sensors to NO2 and OX-B431 sensors to NO2 and

O3 were highly linear over the concentrations studied (R2 ≥ 0.99). The NO2-B43F sensor

did not respond to O3 gas. In mixtures of NO2 and O3, the mean percent bias was between

-8 and 29% for NO2 and between -187 and -24% for O3. We observed changes in senor

baseline over 4 days of experiments equivalent to 34 ppb O3, prompting an alternate

method of baseline-correcting sensor signal to calculate concentrations. The baseline-

correction method resulted in mean percent bias between -44 and 17% for NO2 and

between -107 and 5% for O3. Both analysis methods progressively underestimated O3

concentrations as the ratio of NO2 signal to O3 signal increased. Our results suggest that

paired NO2-B43F and OX-B431 electrochemical sensors permit quantification of O3 in

11

mixture with NO2, but that O3 concentration estimates are less accurate and precise than

those for NO2.

INTRODUCTION

Low-cost sensor networks are playing a profound role in the lower-accuracy/larger

sample measurement paradigm emerging in environmental health (Kumar et al. 2015;

Lewis et al. 2016; Masson et al. 2015; Piedrahita et al. 2014; Snyder et al. 2013). Each

node within such networks is commonly equipped with sensors that produce an electrical

signal proportional to the concentration of a target gas (Kularatna and Sudantha 2008;

Kumar et al. 2011). Reference instruments for gas pollutants commonly utilize

technologies such as optical (UV) spectroscopy (fluorescence, chemiluminescence,

absorption), but these technologies have a number of disadvantages for producing highly-

resolved space-time measurements in the environment, including high initial costs, the

need for skilled operators, and designs geared towards benchtop, laboratory or regulatory

applications (Kularatna and Sudantha 2008; Lewis et al. 2016; Piedrahita et al. 2014;

Snyder et al. 2013). The low cost, small size, portability and low power consumption, of

gas sensors present an opportunity to overcome some of the disadvantages of reference

instruments (Lee and Lee 2001; Masson et al. 2015; Piedrahita et al. 2014; Xiong and

Compton 2014). However, gas sensors require thorough laboratory/field calibration, have

lower sensitivity/specificity, exhibit cross-sensitivity with non-target species, are subject

to signal baseline drift over time and produce data of lower quality (Lewis and Edwards

2016; Lewis et al. 2016; Masson et al. 2015; Piedrahita et al. 2014; Snyder et al. 2013).

12

Electrochemical gas sensors are capable of quantifying a range of target gases including

carbon monoxide, ozone, oxides of nitrogen, hydrogen sulfide, chlorine, and sulfur

dioxide at part-per-million and -billion concentrations (Kumar et al. 2011; Zappi et al.

2012). The principle of operation of electrochemical sensors relies on a chemical

reaction, typically an oxidation or reduction reaction, taking place between an electrode

and the target gas that produces an electrical signal proportional to the gas concentration.

The electrode composition depends on the gas of interest and the chemical reaction that

must take place to detect the target gas (Kumar et al. 2011; Kumar et al. 2013; Mead et

al. 2013; Spinelle et al. 2015a). The reaction creates a difference in electric potential

between the sensor’s working and counter electrodes, which generates an electric current

that constitutes the sensor’s output signal (Kumar et al. 2011; Kumar et al. 2013; Mead et

al. 2013; Spinelle et al. 2015a; Spinelle et al. 2015b). Electrochemical sensors are

typically paired with a potentiostatic circuit, which processes the sensor signal from a

current to a voltage (Kumar and Hancke 2014). In general, they demonstrate sufficient

selectivity for the target gas, high accuracy, linearity, and repeatability, and low power

consumption, making them widely used in many portable direct-reading instruments

(Masson et al. 2015). The disadvantages of electrochemical sensors include electrolyte

loss, a lifespan limited to two years or less (especially in low relative humidity or high

concentration environments), sensitivity to electromagnetic frequencies, and cross-

sensitivity with interfering gases (Kumar et al. 2011; Xiong and Compton 2014).

Although low-cost sensors can be customized for particular applications and

configurations, the need for laboratory set up, calibration and a characterization of cross-

13

sensitivities of electrochemical sensors is well recognized and inhibits their ease of use

(Lewis et al. 2016; Masson et al. 2015; Mead et al. 2013). For example, many existing

commercial electrochemical sensors for O3 and NO2 respond to both gases

simultaneously, without discrimination, due to the fact that NO2 and O3 are both reducible

at similar potentials on carbon and gold electrodes (Hossain et al. 2016). These are in

effect “oxidative gas” sensors, and their response is proportional to the combined

concentration of O3 and NO2. Previous studies have characterized the response of

oxidative gas sensors to their target and interfering gases (Lewis et al. 2016; Spinelle et

al. 2015a), while others have attempted to differentiate sensor response between target

and interfering gas using statistical modelling techniques such as linear regression and

artificial neural networks (Spinelle et al. 2015b).

To address the simultaneous quantification of O3 and NO2 concentrations, Alphasense

Ltd. (Essex, UK) has proposed utilizing a pair of collocated electrochemical sensors: one

that responds to NO2 and O3 (model OX-B431; a typical “oxidative gas” sensor) and one

sensor that only responds to NO2 (model NO2-B43F). The NO2-B43F sensor is equipped

with a manganese dioxide (MnO2) filter which catalyzes O3 into oxygen (O2), thereby

preventing sensor response to O3 in the environment. The response of the oxidative gas

sensor to O3 is calculated by subtracting the response to NO2. This paired sensor method

of quantifying O3 was previously introduced and the differential response of one pair of

sensors was demonstrated for concentrations of 1 ppm O3, 1 ppm NO2 and a mixture of 1

ppm O3 and 1 ppm NO2 (Hossain et al. 2016). The authors however did not evaluate how

well the paired sensor method quantified NO2 and O3 concentrations. We have previously

14

reported on an earlier generation of the OX-B431 sensor (model OX-B421) only, and its

response to NO2 and O3 exclusively (Afshar-Mohajer et al. 2017). In this study, we

assess the bias and precision of the paired sensor method for quantifying NO2 and O3

concentrations at atmospherically-relevant concentrations. The response of the sensors to

NO2 and O3 gas individually was used to create calibration curves and those calibration

curves were used to calculate the concentration of each gas in mixtures over a range of

NO2 and O3 concentrations. We also outline the practical aspects of setting up, calibrating

and using the paired sensor method for quantifying O3.

METHODS

Sensor Configuration

We mounted 3 pairs of new oxidative gas (model OX-B431, Alphasense Ltd., Essex,

UK) and NO2 (model NO2-B43F, Alphasense Ltd., Essex, UK) sensors onto Individual

Sensor Boards (000-0ISB-02) produced by the same manufacturer (Figure 2.1). These

assemblies were connected to a microcontroller (model Seeeduino Cloud, Seeed

Technology Inc., Shenzhen, China) through a customized circuit board, two sensors to

each board. The working pin and reference pin signals were each amplified by a factor of

2 with signal amplifiers (model MCP6002, Microchip Technology Inc., Chandler, AZ)

and fed into a 10-bit analog-to-digital convertor on the microcontroller. The 5-volt power

for the device was smoothed with a 5-volt LM7805 linear regulator to reduce signal

noise. Voltage outputs from each sensor were calculated by taking the difference between

the working and reference pin values and were transmitted over a serial channel

approximately every 2 seconds to a computer.

15

Experimental Setup

The sensors were opened, installed on their assemblies, were run in ambient laboratory

air, underwent preliminary testing with target gases, and had adjustments made to their

supporting hardware and software for approximately 50 days prior to data collection for

this study. Over the course of 4 days, we carried out a series of experiments on three pairs

of OX-B431 and NO2-B43F sensors under different O3 and NO2 concentrations and

mixtures of the two gases. The concentrations of O3 studied were approximately 0, 30,

65, 95 and 125 ppb, and the concentrations of NO2 studied were 0, 0.1, 0.25, 0.5, and 1.0

ppm. These concentrations were reflective of current occupational and environmental

regulatory standards (O3: OSHA Permissible Exposure Limit (PEL) = 0.1 ppm and the

EPA National Ambient Air Quality Standards (NAAQS) = 0.070 ppm; NO2: OSHA PEL

= 5 ppm and the EPA NAAQS = 100 ppb).

We exposed sensors to O3 and NO2 in a 22 cm x 15 cm x 24 cm (7.92 L) acrylic

chamber (Figure 2). A small vent in the chamber allowed gas to escape and the chamber

to operate at a slightly positive pressure with respect to the room. A digital

thermometer/hygrometer (model Hygrochron iButton, Maxim Integrated Inc., San Jose,

CA) monitored the chamber’s temperature and relative humidity. Both O3 and NO2

concentrations in the chamber were measured with highly specific reference instruments

(NO2: model 42c, Thermo Environmental Instruments Inc., Franklin, MA; O3: model

16

Personal Ozone Monitor, “POM,” 2B Technologies Inc., Boulder, CO). The Thermo 42c

chemiluminescent analyzer is a designated federal reference method (FRM) for NO2 and

the POM UV absorption instrument is a designated federal equivalent method (FEM) for

O3. Both instruments were calibrated before use, and experimental conditions were

within the instruments’ operating ranges. Nitrogen dioxide was supplied to the chamber

with a dynamic gas calibrator (model 146i, Thermo Environmental Instruments Inc.,

Franklin, MA) by diluting high-concentration (500 ppm) NO2 from a tank with zero-air.

Ozone was supplied to the chamber by an O3 generator (model 146c, Thermo

Environmental Instruments Inc., Franklin, MA). Airflow from both the dynamic gas

calibrator and ozone generator were supplied to the chamber at 5.0 L/min during all

experiments (including at gas concentrations equal to zero), and concentrations of NO2

and O3 were adjusted to achieve the target gas concentrations in the chamber. We

maintained temperature between 24.6-27.8°C and relative humidity between 36.3-51.8%

by circulating chamber air through a bubbler filled with water at a flowrate of 25 L/min

with a vacuum pump (MEDO VP0435A, Roselle, IL) because we were unable to

condition the air prior to introduction to the chamber.

Although the manufacturer provides calibration slopes and intercepts for each sensor, we

first conducted experiments to develop sensor-specific calibration curves for O3 and NO2

exclusively with the sensor setup and configuration used in this study. We also performed

experiments to assess how well the sensor pairs quantified concentrations of NO2 and O3

in mixtures with one another. For each target concentration of 0.1, 0.25, 0.5 and 1.0 ppm

17

NO2, the chamber was first flushed with zero air for 10 minutes during which the sensor

baseline response was recorded. Then, steady-state NO2 concentration was established for

10 minutes, followed by adding O3 and maintaining concentrations of approximately 0,

65, 125, 30, 95 and 0 ppb for 10 minutes each. A 10-min average of the sensors’ 2-

second voltage output from each experimental condition was used to establish the

response of the sensor to the target concentration(s). Additionally, the standard deviation

of each sensor’s response at each experimental condition was calculated and a mean for

all sensors and experimental conditions was used to characterize the observed sensor

noise.

Calculating Nitrogen Dioxide and Ozone Concentrations with Low-Cost Sensors

According to the manufacturer, OX-B431 sensors are sensitive to both NO2 and O3, and

the total response of the OX-B431 sensors is a sum of the response from NO2 and O3. In

contrast, NO2-B43F sensors respond only to NO2. Consequently, separate calibration

curves for the OX-B431 sensors to NO2, OX-B431 sensors to O3 and NO2-B43F sensors

to NO2 were first determined. Here, calibration curves for each sensor were developed by

applying least-squares linear regression to sensor signal in response to NO2 and O3

exclusively. To measure O3 in mixture with NO2, the NO2-B43F and OX-B431 sensors

must be collocated and the NO2 contribution to the OX-B431 sensor response is

subtracted by first calculating the NO2 concentration with the NO2-B43F sensor. To test

this procedure, we conducted experiments mixing NO2 and O3, and measuring both gases

with electrochemical sensor pairs. We then calculated NO2 and O3 concentrations for

18

experimental conditions using two different methods in response to an observed change

in sensor baseline values over the course of experiments.

Method 1: Applying Calibration Slope and Intercept

In the first analysis, subsequently referred to as “Method 1,” the slopes and intercepts of

each of the sensors determined by our calibration experiments with a single gas were

applied to the sensor response. The calibration curve derived for the NO2-B43F

(Equation 1) was rearranged to solve for the NO2 concentration (Equation 2):

𝑚𝑉𝑁𝑂2−𝐵43𝐹 = [𝑁𝑂2]𝑚𝑁𝑂2−𝐵43𝐹 + 𝑏𝑁𝑂2−𝐵43𝐹 (1)

[𝑁𝑂2] =𝑚𝑉𝑁𝑂2−𝐵43𝐹−𝑏𝑁𝑂2−𝐵43𝐹

𝑚𝑁𝑂2−𝐵43𝐹 (2)

where mVNO2-B43F is the response of the NO2-B43F sensor in millivolts (mV), [NO2] is

the concentration of NO2, mNO2-B43F is the slope of the calibration curve of the NO2-B43F

sensor, and bNO2-B43F is the intercept of the calibration curve for the NO2-B43F sensor. In

Method 1, we approached the signal from the OX-B431 sensor in a similar fashion as the

NO2-B43F sensor, including terms for the OX-B431 sensor calibration slope to NO2 and

O3 and the calibration intercept for O3 (Equation 3):

𝑚𝑉𝑂𝑋−𝐵431 = [𝑁𝑂2]𝑚𝑂𝑋−𝐵431,𝑁𝑂2+ [𝑂3]𝑚𝑂𝑋−𝐵431,𝑂3

+ 𝑏𝑂𝑋−𝐵431 (3)

19

where mVOX-B431 is the response of the OX-B431 sensor in mV, mOX-B431,NO2 is the slope

of the calibration curve of the OX-B431 sensor to NO2, [O3] is the concentration of O3,

mOX-B431,O3 is the slope of the OX-B431 sensor to O3 and bOX-B431 is the intercept of the

OX-B431 sensor determined in the OX-B431 sensor O3 calibration.

We observed variability in sensor intercepts in NO2 and O3 calibration experiments and

chose to use the O3 calibration intercept because O3 was the gas of interest. The

contribution of NO2 gas to the OX-B431 sensor response can be inferred with the OX-

B431 senor NO2 gas slope, mOX-B431,NO2, and the concentration of NO2 estimated from the

NO2-B43F sensor. To solve for the concentration of O3, both the intercept of the OX-

B431 sensor and the contribution of NO2 gas to the OX-B431 sensor response was

subtracted from the total OX-B431 sensor response and divided by the slope of the OX-

B431 O3 gas calibration curve (Equation 4).

[𝑂3] =𝑚𝑉𝑂𝑋−𝐵431−𝑏𝑂𝑋−𝐵431−[𝑁𝑂2]𝑚𝑂𝑋−𝐵431,𝑁𝑂2

𝑚𝑂𝑋−𝐵431,𝑂3 (4)

Method 2: Baseline-Correcting Sensor Response and Applying Calibration Slope

In an alternate analysis, prompted by an observed change in sensor baseline values and

subsequently referred to as “Method 2,” we applied only the slopes determined in the

calibration experiments to the baseline-corrected sensor response. The baseline response

for each sensor was recorded at a concentration of zero ppm NO2 and zero ppb O3 at the

beginning of each of the NO2 and O3 mixture experiments and subtracted from all the

20

sensor’s readings in the experiment, thus eliminating the need for a sensor intercept. The

relationship between sensors’ response and gas concentrations were therefore:

𝑚𝑉𝑁𝑂2−𝐵43𝐹,𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒−𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑 = [𝑁𝑂2]𝑚𝑁𝑂2−𝐵43𝐹,𝑁𝑂2 (5)

and

𝑚𝑉𝑂𝑋−𝐵431,𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒−𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑 = [𝑁𝑂2]𝑚𝑂𝑋−𝐵431,𝑁𝑂2+ [𝑂3]𝑚𝑂𝑋−𝐵431,𝑂3

(6)

where mVNO2-B43F,baseline-corrected and mVOX-B431,baseline-corrected are baseline-corrected signals

from the NO2-B4F and OX-B431 sensors, respectively, and all other terms remain the

same. Method 2 provides a strategy to manage transient changes in sensor baseline,

which is comparable to the calibration intercept, but assumes that sensor calibration slope

is constant for the dataset.

Bias and Precision of NO2 and O3 Concentrations Estimated by Electrochemical

Sensors

To quantify the accuracy of sensor concentration estimates, measurement error was taken

as the percent bias of each NO2 and O3 concentration estimate for each sensor pair and

the average of the 3 sensor pairs compared to the reference instruments. Bias was

calculated according to:

21

%𝐵𝑖𝑎𝑠 = 𝑆𝑒𝑛𝑠𝑜𝑟−𝑅𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒

𝑅𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒× 100% (7)

where Sensor is the concentration estimated from the electrochemical sensors and

Reference is the concentration according to the reference instruments. We estimated the

concentration of NO2 and O3 using each sensor pair and then took the mean concentration

of the 3 sensor pairs. This mean concentration estimate was evaluated against the

reference instruments to calculate the mean bias of NO2 and O3 concentration estimates

for each experimental condition. Bias was compared to guidance values from NIOSH and

the EPA for direct reading monitors and air sensors. NIOSH specifies that percent bias

should be within ± 10% (NIOSH 2012), whereas the EPA recommends that bias be

within 20 to 50%, depending on the application area, including Education and

Information (< 50%), Hotspot Identification and Characterization (< 30%) Supplemental

Monitoring (< 20%), and Personal exposure (< 30%). (2014) Similarly, the mean

absolute percent error (MAPE) was calculated to summarize the measurement error of

NO2 and O3 concentration estimates of mixture experiments for each sensor pair and the

average of 3 sensor pairs, according to:

𝑀𝐴𝑃𝐸 = 100%

𝑛× ∑ |

𝑆𝑒𝑛𝑠𝑜𝑟−𝑅𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒

𝑅𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒|𝑛

𝑖=1 (8)

where n is the number of NO2 or O3 experimental concentrations studied, Sensor is the

NO2 or O3 concentration measured by the electrochemical sensors and Reference is the

O3 or NO2 concentration measured by the reference instrument. Percent bias and MAPE

22

were not calculated for the lowest concentrations of NO2 and O3 where the target

concentration was zero. Here, bias provides a measure of error at each experimental

condition within mixture experiments, and MAPE summarizes the bias observed across

the range of conditions for each mixture experiment.

To characterize the precision of gas concentration estimates, we calculated the coefficient

of variation, CV, by dividing the standard deviation of the 3 senor concentration

estimates, σ, by the absolute value of the mean concentration estimate of the sensors, |μ|,

and expressed it as a percent:

𝐶𝑉 = 𝜎

|𝜇|× 100% (9)

Taking the absolute value of the mean of sensor concentration estimates allowed for the

calculation of precision when sensor signals produced concentration estimates that were

negative. Negative estimates of gas concentration are an artifact of processing sensor

signal to gas concentration, particularly at low concentrations. Higher coefficient of

variation indicates more variability and more imprecision in concentration estimates. We

compared the coefficient of variation to guidance values from the EPA which

recommends precision between 20 and 50% depending on the application (EPA 2014).

NIOSH does not provide a recommended value for precision. All data were analyzed

with MATLAB R2017a (Natick, MA).

23

RESULTS

Sensor Response to Nitrogen Dioxide or Ozone Exclusively

The results of the linear regression on voltage output from each sensor with respect to

NO2 and O3 exclusively are shown in Table 2.1. Among the 3 NO2-B43F sensors, the

mean of the slopes of the response observed to NO2 was 283 mV/ppm with a standard

deviation of 27 mV/ppm (9% of the mean). Individual sensors’ response to NO2 was

highly statistically significant (p < 0.00001) and linear (R2 = 1.00). In contrast, the slopes

of the response of the NO2-B43F sensors to O3 were low (mean slope = 13 mV/ppm), not

statistically significant (p ≥ 0.165), and non-linear (R2 ≥ 0.09), consistent with the

expectation that these sensors do not respond to O3.

The mean slope of the response of the OX-B431 sensors to NO2 was 382 mV/ppm and to

O3 was 431 mV/ppm. The standard deviations of the mean slopes were 56 mV/ppm NO2

(15% of the mean) and 80 mV/ppm for O3 (19% of the mean). Individual OX-

B431sensor response to NO2 was highly statistically significant (p < 0.00001) and linear

(R2 = 1.00). Similarly, the individual OX-B431 sensor response to O3 was highly

statistically significant (p < 0.0005) and linear (R2 = 0.99). Of note, the mean OX-B431

sensor response was 1.35-times larger than to NO2 gas than the NO2-B43F sensors (382

versus 283 mV/ppm) and the mean OX-B431 sensor response to O3 was 1.13-times

greater than to NO2 on a concentration basis (431 mV/ppm versus 382 mV/ppm).

Calculating NO2 and O3 Concentrations by Applying Calibration Slope and

Intercept: Method 1

24

The mean bias of NO2 and O3 concentration estimates for Method 1 are shown in Figure

2.3, Panels (a) and (c). The mean bias of NO2 and O3 concentration estimates for each

experimental condition was calculated using the mean gas concentration estimate of the 3

sensor pairs. The mean bias points shown in Figure 2.3 are colored based on this value.

The contour plot was created by linear interpolation of the overlying mean bias points to

describe the bias between experimental conditions. Bias is an indicator of accuracy and

values closer to zero represent closer agreement of the electrochemical sensors to the

reference instrument. For Method 1, the mean bias for of NO2 ranged from -8 to 29%,

with bias of a larger magnitude observed at higher NO2 concentrations (Figure 2.3a). The

mean O3 bias was between -187 and -24% with higher bias (greater underestimation)

observed at lower O3 concentrations (Figure 2.3c). For Method 1, 17 out of 20 (85%)

NO2 concentration estimates and zero out of 16 O3 concentration estimates met the

NIOSH criterion of bias ± 10%. The bias of individual sensor pair concentration

estimates of NO2 and O3 for Method 1 are presented in Appendix 2.1. The MAPE is

interpreted here as a summary measure of experimental biases and shown in Table 2.3.

For NO2 concentration estimates, the overall MAPE was equal to 8%, less than NIOSH’s

bias criterion of ± 10% (Table 2.3). For O3 concentration estimates, the overall MAPE

was equal to 71% (Table 2.3) and was greater than the largest EPA criterion for bias of ±

50%.

The mean variation in NO2 and O3 concentrations estimated with the 3 sensor pairs at

each experimental condition for Method 1 is shown in Figure 2.3, Panels (b) and (d). We

observed generally uniform variation in concentration estimates between 1 and 7%

25

(median = 5%) for NO2 (Figure 2.3b), but strongly increasing variation in ozone ranging

from 6 to 146% (median = 44%) for O3 that increased with increasing NO2 and

decreasing O3 concentrations (Figure 2.3d). The NO2 concentrations estimated via

Method 1 met the most stringent EPA guidelines for precision (< 20%), whereas 6 out of

16 (38%) of the O3 concentration estimates met the same guideline.

Calculating NO2 and O3 Concentrations by Applying Calibration Slope to Baseline-

Corrected Sensor Response: Method 2

We observed decreases in sensor response to zero air (zero ppm NO2 and zero ppb O3)

ranging from 12 to 22 mV over the 4 days of experiments that were unrelated to

temperature or relative humidity differences (Table 2.2). These baseline voltages

decreased over the 4 days by as much as 107% for the OX-B431 and 92% for the NO2-

B43F sensors comparing the first day of experiments to the last day. Of particular note,

was the observed change in sensor baseline compared to the magnitude of the sensor

response to target gas. For example, the OX-B431 sensor with the largest absolute change

in sensor baseline among the OX-B431 sensors had a change of 17 mV, corresponding to

approximately 0.040 ppm NO2 or 0.034 ppm (~34 ppb) O3. The NO2-B43F sensor with

the greatest absolute change in sensor baseline among the NO2-B43F sensors had a

change of 22 mV, corresponding to approximately 0.070 ppm NO2.

For Method 2, we observed higher levels of bias for NO2 concentration estimates but

lower levels of bias for O3 concentration estimates compared to Method 1 (Figure 2.4).

For Method 2, the mean bias for NO2 ranged from -44 to 17%, with the magnitude of the

26

bias higher at lower NO2 concentrations (Figure 2.4a). For O3 concentration estimates,

the mean bias for Method 2 was between -107 and 5% and displayed a pattern of bias

similar to Method 1 with higher bias observed at lower O3 and higher NO2 concentrations

(Figure 2.4c). For Method 2, 10 out of 20 (50%) NO2 concentration estimates and 2 out

of 16 (13%) O3 concentration estimates met the NIOSH criterion of bias ± 10%. The bias

of individual sensor pair concentration estimates of NO2 and O3 for Method 2 are

presented in Appendix 2.2. The overall MAPE of NO2 concentration estimates was equal

to 14% and of O3 concentration estimates was 30% (Table 2.3), which were greater than

the NIOSH criterion of 10%, but for NO2 within the most stringent limits suggested by

the EPA for supplemental monitoring activities (± 20%) and hotspot identification and

characterization (±30%). Variation for Method 2 was calculated the same way as in

Method 1, and we observed comparable variation in concentration estimates between 2

and 6% (median = 5%) for NO2 (Figure 2.4b). For O3, variation was between 3 and

1753% (median = 20%) and increased with increasing NO2 and decreasing O3

concentrations (Figure 2.4d). NO2 concentrations estimated via Method 2 met the most

stringent EPA guidelines for precision (< 20%), whereas 9 out of 16 (56%) of the O3

concentration estimates met the same guideline.

For Method 1 and Method 2 overall, the bias and variation of NO2 concentration

estimates were less than for O3 (Figure 2.3 and Figure 2.4), indicating that concentration

estimates of NO2 were more accurate and precise compared to those for O3. We observed

a larger overall error of O3 concentration estimates for Method 1 compared to Method 2

(MAPE = 71 versus 30%) and a larger overall error of NO2 concentration estimates for

Method 2 compared to Method 1 (MAPE = 14 versus 8%) (Table 2.3). Method 1

27

produced NO2 concentration estimates with MAPE between 3 and 12% with generally

smaller error at low NO2 concentrations, and Method 2 produced concentration estimates

between 4 and 34% with smaller error at high NO2 concentrations (Table 2.3). In

addition, while the variation of concentration estimates with both methods was

comparable for NO2, for O3, the variation increased as the concentration of NO2 increased

and O3 decreased and was generally smaller for Method 2 excepting one outlier. (Figure

2.3d and Figure 2.4d). Throughout all experiments we observed a level of noise for all

sensors, characterized as the mean standard deviation of the sensor signal at each steady-

state condition, equal to 5.0 mV (0.1% full scale), which was approximately equivalent to

0.02 ppm NO2 or 12 ppb O3.

DISCUSSION

Our calibration experiments over 0-1.0 ppm NO2 and 0-125 ppb O3 demonstrated that the

NO2-B43F sensors had a highly linear response to NO2, and that the OX-B431 sensors

had a highly linear response to NO2 and O3, comparable with previous studies (Afshar-

Mohajer et al. 2017). The variability we observed in the calibration slope among the 3

sensors of each type (Table 2.1) is consistent with the variability in the sensor-specific

calibration slopes provided by the manufacturer. In this regard, in our sample of 3 sensors

of each type, the standard deviation of the mean of the calibration slopes among the 3

NO2-B43F sensors to NO2 was equal to 27 mV/ppm, and for the OX-B431 sensors to O3

was equal to 81 mV/ppm, and to NO2 was equal to 56 mV/ppm. These results suggest it is

appropriate to use sensor-specific calibration curves rather than a common curve for each

type of sensor. However, we caution against generalizing the variability in calibration

28

slope observed here to the whole population of these sensors, due to the small size of our

sample.

We tested the ability of collocated pairs of electrochemical sensors to quantify NO2 and

O3 concentrations in mixture over a range of concentrations of both gases and observed

this strategy works, although with decreasing accuracy and precision when the signal

from NO2 obscures or swamps the signal from O3. Even though our sample of 3 sensors

of each type was small, we expect this characteristic is true with a larger sample of

sensors also, given the trend held for each sensor pair in this study, albeit to varying

degrees. On an individual sensor pair level, the accuracy of NO2 and O3 concentration

estimates varied across the 3 sensor pairs studied, with one of the sensor pairs out-

performing the other two according to the MAPE. This suggests that sensor pairs should

be calibrated and tested in the laboratory prior to deployment in the field to identify

sensor pairs with unacceptable levels of measurement error.

In this series of experiments, we observed the specificity of the NO2-B43F sensor. The

NO2-B43F sensor is similar to the OX-B431 sensor but is fitted with a magnesium

dioxide filter that prevents O3 from reaching the sensor electrode (Hossain et al. 2016).

Although we observed the NO2-B43F sensors respond slightly to increasing

concentrations of O3 gas, on average, their response to NO2 gas was over 20-times greater

than the response to O3. Furthermore, least-squares regression of the response among the

3 NO2-B43F sensors to O3 produced p-values 0.16 ≤ p ≤ 0.63, indicating that O3

concentration was not a significant predictor of NO2-B43F sensor response. These results

29

provide evidence that the MnO2 filter on the NO2-B43F sensor is effective at excluding

O3 from the sensor under the range of concentrations studied and are consistent with prior

evaluations of excluding O3 with an MnO2 filter (Hossain et al. 2016).

In this study we identified an important source of measurement error: changes in the

baseline responses of sensors. Over the course of the 4 days in which we carried out

calibration experiments and NO2 and O3 mixture experiments we observed changes in

sensor baseline that affected the quantification of NO2 and O3 and the measurement error

associated with each sensor concentration estimate. Compared to the NO2-B43F sensor, a

unit of the OX-B431 sensor signal is associated with a greater concentration of gas,

making concentration estimates more vulnerable to errors given a change in sensor

baseline. This is an especially acute problem given that O3 concentrations of interest are

often less than 100 ppb, compared to NO2 concentrations which are often greater than

100 ppb. Our strategy to correct for changes in sensor baseline resulted in a differential

change in error associated with NO2 and O3 concentration estimates between the two

methods. For Method 1 compared to Method 2, we observed a higher degree of accuracy

in NO2 concentration estimates (MAPE = 8 versus 14%), but worse accuracy for O3

concentration estimates (MAPE = 71 versus 30%). In this laboratory study were easily

able to accommodate changes in sensor baseline with Method 2, however, a comparable

methodology in the field on the day-to-day timescale may be impractical.

Here we demonstrate there is more error estimating O3 concentration in a mixture with

NO2 with paired electrochemical sensors compared to estimating NO2 concentration with

30

a single sensor because the error associated with 2 sensors is propagated through the

subtraction procedure. Additionally, if using a common sensor calibration slope for

sensors of the same type, it may be difficult to quantify the concentration of O3 in a

mixture with NO2 because the mean response of the OX-B431 sensors to O3 may be

smaller in magnitude than the variability of the OX-B431sensors’ response to NO2. For

example, we observed the range of response across the 3 OX-B431 sensors exposed to

0.5 ppm NO2 equal to 57 mV which is equivalent to the mean OX-B431 sensor response

to 132 ppb O3. Another challenge is that the changes in sensor baseline are large relative

compared to the response of the sensor to O3 at typical ambient and occupational

concentrations. Here we observed maximum changes of 17 mV with the OX-B431 senor

associated with 34 ppb O3 and 22 mV with the NO2-B43F associated with 0.070 ppm

NO2. These dynamics make accurate O3 concentration estimates in a mixture with NO2

challenging with pairs of electrochemical sensors. This is particularly relevant if the end-

user chooses to use a common calibration curve for sensors of each type as opposed to

individual calibration curves for each sensor. For these reasons, when measuring O3

concentrations with paired electrochemical sensors, we caution against using single

calibration curves for each sensor type without previously examining individual sensor

response to target gas. This conclusion may not be consistent with previous evaluations

of an earlier-generation oxidative gas sensor (model: OX-B421, Alphasense Ltd., Essex,

UK) where a single calibration curve for NO2 and a single calibration curve for O3

adequately characterized the response of a sample of 3 sensors (Afshar-Mohajer et al.

2017).

31

Our evaluation of these sensors occurred over a stable and controlled range of

temperature and relative humidity for each experimental condition (mean temperature ±

SD: 27 ± 1°C, mean relative humidity ± SD: 39 ± 5%RH). This was intentional because a

well-known trait of electrochemical sensors is that their response is affected by these

parameters, especially sudden changes and we sought to reduce the influence of these

physical parameters on sensors response. Purposefully characterizing sensor response

under a larger range of temperature and relative humidity or applying temperature and

relative humidity correction factors from the manufacturer would be particularly

important for deployment in environments where temperature and relative humidity are

highly variable.

A limitation of this study is that we did not examine other gases that interfere with the

quantification of O3 concentration with a pair of electrochemical sensors. Species such as

nitrogen monoxide (NO) and carbon dioxide (CO2), are strong interferents for NO2 and

oxidative gas sensors and have the potential to substantially impact the concentration

estimates of target gases. In a study using previous generations of the sensors used here at

ambient concentrations of CO2, NO, NO2 and O3, the impact observed on O3

concentration estimates by the OX-B421 sensor was 20.6% for NO and 365.8% for CO2,

whereas the impact on NO2 concentration estimates by the NO2-B4 sensor (Alphasense

Ltd., Essex, UK) was -20.6% for NO and 118.9% for CO2 (Lewis et al. 2016). These

gases co-occur with O3 and NO2 in ambient and occupational environments and would

decrease the accuracy of concentration estimates, or may completely swamp target gas

signals if present in high concentrations. The present study with NO2 and O3 provides

32

evidence that the strategy of filtering out cross sensitive gases and deploying collocated

sensors could be successfully developed for other target gases with known interferents

depending on the required accuracy of the application.

CONCLUSIONS

We evaluated a method for measuring NO2 and O3 in mixture using paired

electrochemical sensors: one sensor that responds to O3 and NO2 (OX-B431) and another

that responds to NO2 only (NO2-B43F). We observed the strategy works over a range of

concentrations and mixtures of the two gases, but the precision and accuracy of O3

concentration estimates declined as NO2 concentration increased. We observed

substantial variability in the concentration estimates of O3 in a sample of 3 sensor pairs.

Over the course of the 4 days of experiments, we also observed a change in senor

baseline, complicating the calculation of O3, and prompting an alternate method of

calculating concentration from sensor signal. Although the paired sensor method has

potential to improve the specificity of O3 concentration estimates compared to a single

oxidative gas sensor, concentrations of NO2 and O3 where the ratio of NO2 signal to O3

signal is large may still challenge their performance, performance among sensor pairs is

variable, sensor baseline voltage is subject to drift and the cost to measure O3 effectively

doubles. Increases in target gas specificity will ameliorate a major drawback and improve

the utility of electrochemical sensors and has the potential to provide higher-quality data

for environmental and occupational sensor networks.

33

TABLES & FIGURES

Table 2.1. Summary of Least-Squares Linear Regression.

NO2 Calibration O3 Calibration

Sensor Pair

Slope

(Est ± SE)

mV/ppm

Intercept

(Est ± SE)

mV

R2

Slope

(Est ± SE)

mV/ppm

Intercept

(Est ± SE)

mV

R2

a) NO2-B43FSensors

1 282.0B ± 1.3 7.4 ± 0.7 1.00 7.8 ± 14.8 14.5 ± 1.2 0.09

2 256.4B ± 1.6 -2.4 ± 0.9 1.00 10.0 ± 15.6 3.9 ± 1.2 0.12

3 309.5B ± 4.8 2.0 ± 2.6 1.00 22.2 ± 12.2 11.8 ± 1.0 0.53

Mean (SD) 282.7 (26.6) 2.3 (4.9) -- 13.4 (7.8) 10.1 (5.5) --

b) OX-B431 Sensor

1 376.6B ± 3.2 5.5 ± 1.7 1.00 424.3A ± 25.5 13.2 ± 2.0 0.99

2 328.1B ± 2.7 -5.7 ± 1.5 1.00 354.5A ± 21.2 2.3 ± 1.7 0.99

3 440.0B ± 3.5 8.7 ± 1.9 1.00 515.3A ± 27.4 17.4 ± 2.2 0.99

Mean (SD) 381.5 (56.1) 2.8 (7.6) -- 431.4 (80.6) 11.0 (7.8) --

Notes: SE: standard error of the regression A least-squares regression coefficient of sensor slope with p-value < 0.0005 B least-squares regression coefficient of sensor slope with p-value < 0.00001

34

Table 2.2. Sensor Response (mV) to Zero Air and Temperature and Relative Humidity of Zero Air During Baseline.

Temp ± SD

(°C)

RH ± SD

(%)

NO2-B43F Sensor OX-B431 Sensor

1 2 3 1 2 3

Day 1 AM 23.4 ± 0.3 25.9 ± 1.7 23.0 12.8 22.7 23.8 14.2 31.8

Day 1 PM 27.7 ± 0.0 44.0 ± 0.4 19.4 9.9 19.1 14.1 9.4 24.5

Day 2 AM 23.7 ± 0.3 26.9 ± 5.9 18.7 8.2 19.0 20.3 10.8 25.6

Day 2 PM 27.7 ± 0.0 42.5 ± 2.6 17.0 6.4 16.6 17.2 6.1 23.3

Day 4 AM 23.6 ± 0.2 48.1 ± 0.6 13.4 2.9 10.5 10.5 0.1 14.6

Day 4 PM 26.2 ± 0.2 46.8 ± 0.4 10.1 1.1 0.8 10.5 -1.1 14.4

Average ± SD 25.3 ± 2.1 39.0 ± 10.0 18.3 ± 4.6 6.9 ± 4.4 14.8 ± 7.9 17.2 ± 5.4 8.1 ± 6.1 24.0 ± 6.8

Change, mV (%) -- -- -12.9 (-56) -11.7 (-92) -21.8 (-96) -13.3 (-56) -15.3 (-107) -17.4 (-55)

Notes: SD: standard deviation

35

Table 2.3. MAPE for NO2 and O3 Averaged for 3 Sensor Pairs.

Method 1 Method 2

Experiment MAPE [NO2]

(%)

MAPE [O3]

(%)

MAPE [NO2]

(%)

MAPE [O3]

(%)

0.1 ppm NO2 3 47 34 11

0.25 ppm NO2 6 57 14 23

0.5 ppm NO2 12 87 5 36

1.0 ppm NO2 10 91 4 51

Overall Mean 8 71 14 30

Notes: MAPE: mean absolute percent error

36

Figure 2.1. Sensor setup. Oxidative gas and NO2 sensors were mounted onto

Individual Sensor Boards. A custom circuit board connected the sensor-ISB

assembly to a microcontroller.

37

Figure 2.2. Setup used for the calibration of NO2-B43F and OX-B431 sensors and

experiments exposing paired electrochemical sensors to mixtures of O3 and NO2.

38

Figure 2.3. Bias and coefficient of variation maps of NO2 and O3 concentration

estimates for Method 1. Variation was calculated as the standard deviation of the

concentration estimates of 3 sensor pairs divided by the absolute value of the mean

estimate.

39

Figure 2.4. Bias and coefficient of variation maps of NO2 and O3 concentration

estimates for Method 2. Variation was calculated as the standard deviation of the

concentration estimates of 3 sensor pairs divided by the absolute value of the mean

estimate.

40

REFERENCES

Afshar-Mohajer N, Zuidema C, Sousan S, Hallett L, Tatum M, Rule AM, et al. 2017.

Evaluation of low-cost electro-chemical sensors for environmental monitoring of

ozone, nitrogen dioxide and carbon monoxide. Journal of Occupational and

Environmental Hygiene:0-0.

EPA 2014. Air sensor guidebook. EPA 600/R-14/159.Office of Research and

Development.

Hossain M, Saffell J, Baron R. 2016. Differentiating no2 and o3 at low cost air quality

amperometric gas sensors. ACS Sensors 1:1291-1294.

Kularatna N, Sudantha BH. 2008. An environmental air pollution monitoring system

based on the ieee 1451 standard for low cost requirements. IEEE Sensors Journal

8:415-422.

Kumar A, Singh IP, Sud SK. 2011. Energy efficient and low-cost indoor environment

monitoring system based on the ieee 1451 standard. IEEE Sensors Journal 11:2598-

2610.

Kumar A, Kim H, Hancke GP. 2013. Environmental monitoring systems: A review.

IEEE Sensors Journal 13:1329-1339.

Kumar A, Hancke GP. 2014. Energy efficient environment monitoring system based on

the ieee 802.15. 4 standard for low cost requirements. IEEE Sensors Journal 14:2557-

2566.

Kumar P, Morawska L, Martani C, Biskos G, Neophytou M, Di Sabatino S, et al. 2015.

The rise of low-cost sensing for managing air pollution in cities. Environment

international 75:199-205.

41

Lee D-D, Lee D-S. 2001. Environmental gas sensors. IEEE Sensors Journal 1:214-224.







Mead MI, Popoola OAM, Stewart GB, Landshoff P, Calleja M, Hayes M, et al. 2013.

The use of electrochemical sensors for monitoring urban air quality in low-cost, high-

density networks. Atmospheric Environment 70:186-203.

NIOSH 2012. Components for evaluation of direct-reading monitors for gases and

vapors. Publication No. 2012-162.Centers for Disease Control and Prevention (CDC)

National Institute for Occupational Safety and Health (NIOSH).







Spinelle L, Gerboles M, Aleixandre M. 2015a. Performance evaluation of amperometric

sensors for the monitoring of o 3 and no 2 in ambient air at ppb level. Procedia

Engineering 120:480-483.

42

Spinelle L, Gerboles M, Villani MG, Aleixandre M, Bonavitacola F. 2015b. Field

calibration of a cluster of low-cost available sensors for air quality monitoring. Part a:

Ozone and nitrogen dioxide. Sensors and Actuators B: Chemical 215:249-257.

Xiong L, Compton RG. 2014. Amperometric gas detection: A review.

IntJElectrochemSci 9:7152-7181.

Zappi P, Bales E, Park JH, Griswold W, Rosing TŠ. The citisense air quality monitoring

mobile sensor node. In: Proceedings of the Proceedings of the 11th ACM/IEEE

Conference on Information Processing in Sensor Networks, Beijing, China, 2012

2012.

43

CHAPTER THREE

Manuscript 2

(Submitted for publication in the Annals of Work Exposure and Health)

Mapping Occupational Hazards with a Multi-Sensor Network

in a Heavy-Vehicle Manufacturing Facility

Christopher Zuidema, Sinan Sousan, Larissa V Stebounova, Alyson Gray, Xiaoxing Liu,

Marcus Tatum, Oliver Stroh, Geb Thomas, Thomas Peters and Kirsten Koehler

44

ABSTRACT

Due to their small size, low power demands and customizability, low-cost sensors can be

deployed in collections that are spatially distributed in the environment, known as sensor

networks. The literature contains examples of such networks in the ambient environment;

this work describes the development and deployment of a 40-node multi-hazard network,

constructed with low-cost sensors for particulate matter (SHARP GP2Y1010AU0F),

carbon monoxide (Alphasense CO-B4), oxidizing gases (Alphasense OX-B431) and

noise (developed in-house) in a heavy-vehicle manufacturing facility. Network nodes

communicated wirelessly with a central database in order to record hazard measurements

at 5-minute intervals. Here, we report on the space-time measurements from the network,

precision of network measurements, and accuracy of network measurements with respect

to field reference instruments through 5 months of continuous deployment. During

typical production periods, 1-hr mean hazard levels ± standard deviation across all

monitors for particulate matter, carbon monoxide, oxidizing gases and noise were 0.52 ±

0.1 mg/m3, 7 ± 2 ppm, 125 ± 27 ppb, and 83 ± 1 dBA respectively. We observed clear

diurnal and weekly temporal patterns for all hazards and daily, hazard-specific spatial

patterns attributable to general manufacturing processes in the facility. Processes

associated with the highest hazard levels were flame cutting (particulate matter), manual

welding and robotic welding (carbon monoxide), machining and welding (oxidizing

gases and noise). Network sensors exhibited varying degrees of precision with 95% of

measurements among 3 collocated nodes within 0.23 mg/m3 for particulate matter, 0.4

ppm for carbon monoxide, 7 ppb for oxidizing gases, and 1 dBA for noise of each other.

The median percent bias with reference to direct-reading instruments was 41%, 7%, 36%

45

and 1%, for particulate matter, carbon monoxide, oxidizing gases and noise respectively.

This study demonstrates the successful long-term deployment of a multi-hazard sensor

network in an industrial manufacturing setting and illustrates the high temporal and

spatial resolution of hazard data that sensor and monitor networks are capable of. We

show that network-derived hazard measurements offer rich datasets to comprehensively

assess occupational hazards. Our network sets the stage for the characterization of

occupational exposures on the individual level with wireless sensor networks.

INTRODUCTION

Low-cost sensors have attracted the attention of environmental health scientists interested

in measuring air pollution with a high degree of spatial and temporal resolution, despite

the sensors’ lower accuracy, precision, sensitivity and specificity (Kumar et al. 2015;

Lewis et al. 2016; Masson et al. 2015; Piedrahita et al. 2014; Snyder et al. 2013). Current

sensor availability reflects regulatory and health priorities (Lewis and Edwards 2016); for

instance, many low-cost sensors are available for particulate matter (PM) (Jovašević-

Stojanović et al. 2015; Sousan et al. 2016) and hazardous gases, such as carbon monoxide

(CO), ozone (O3), nitrogen dioxide (NO2) and sulfur dioxide (SO2) (Xiong and Compton

2014). In the current study, a sensor network was developed to measure PM, CO, O3, and

noise, agents that are important for worker health in industrial manufacturing facilities.

PM has well-established relationships with cardiopulmonary and respiratory diseases,

lung cancer, inflammation, oxidizing stress, pulmonary infection, and lung function

(Anderson et al. 2012; Dockery 1993; Pope et al. 1995; Pope III and Dockery 2006). The

Permissible Exposure Limit (PEL) for respirable PM is 5 mg/m3 (OSHA, 1993a). The

46

health effects of CO at or below the PEL, which is equal to 50 ppm (OSHA, 1993a),

include headache, dizziness, weakness, nausea and confusion (Raub et al. 2000). Ozone

is a well-known oxidant and its inhalation causes inflammation, reduced lung function,

DNA damage and increased symptoms and development of asthma (Bornholdt et al.

2002; Kampa and Castanas 2008; Lippmann 1989; Weschler 2006). The PEL for O3 is

100 ppb (OSHA, 1993). Occupational noise exposure induces hearing impairment,

hypertension and annoyance and may be associated with biochemical effects, immune

effects and changes in absentee rate and performance (Passchier-Vermeer and Passchier

2000). The permissible exposure to noise for an 8-hr work period is 90 dBA (OSHA,

1974).

The standard technique for quantifying PM in the occupational environment is

gravimetric analysis, which requires filters, pumps, and an analytical balance (NIOSH

2017), or a third party to weigh filters. Although a variety of strategies are employed to

sample for hazardous gases in the workplace, such as detector tubes, whole air sampling,

sorbent sampling, and direct-reading instruments (DRIs), each has advantages and

disadvantages (Harper 2004). Major disadvantages of these strategies include the need for

trained professionals, equipment requirements, sample handling, high cost, large

size/weight (Harper 2004). Additionally, for both PM and gases, a major drawback of

these sampling strategies (except for DRIs) is that measurements are time integrated—

commonly 8-10 hours—reflecting typical work shifts. Sensors can overcome some of the

disadvantages of traditional methods because of their low cost, small size, high temporal

resolution, portability and low power consumption (Lee and Lee 2001; Xiong and

47

Compton 2014). However, the drawbacks of sensors include the need for thorough

laboratory/field calibration, lower accuracy, precision, sensitivity and specificity, cross-

sensitivity with non-target species, and instability over time compared with traditional

techniques (Lewis and Edwards 2016; Lewis et al. 2016; Mead et al. 2013). Despite these

challenges, sensors offer a complementary strategy to study human exposure to

occupational and environmental hazards.

Recent advances in open software toolkits and the modularization and commoditization

of microprocessor platforms have facilitated the development of customized wireless

sensor networks applications. Sensor networks are collections of small, inexpensive

devices distributed throughout an environment (Heidemann and Bulusu 2001). A

growing number of examples of environmental health sensor networks (English et al.

2017; Gao et al. 2015; Hasenfratz et al. 2015; Heimann et al. 2015; Ikram et al. 2012;

Jiang et al. 2016; Jiao et al. 2016; Kumar et al. 2011; Mead et al. 2013; Moltchanov et al.

2015) offer potentially powerful tools for hazard mapping, a technique that displays

measured hazard(s) throughout a facility or geographic area (Koehler and Volckens 2011;

Koehler and Peters 2013; O'Brien 2003; Peters et al. 2006). Hazard maps can be used to

visually communicate risk (Koehler and Volckens 2011), identify hazard sources (Evans

et al. 2008; O'Brien 2003), characterize the distribution of hazards in a facility or the

environment (Evans et al. 2008; Ott et al. 2008; Peters et al. 2006), and inform hazard

control strategies (O'Brien 2003). While aerosol mapping in particular has been

successful in a variety of occupational and ambient settings (Evans et al. 2008; Heitbrink

et al. 2007; Liu and Hammond 2010; O'Brien 2003; Ott et al. 2008; Park et al. 2010;

48

Peters et al. 2006; Peters et al. 2012; Vosburgh et al. 2011), there is great potential for

mapping other hazards such as noise, vibration, radiation, gasses, and vapors (Koehler

and Volckens 2011).

Hazard mapping ideally uses frequent measurements at high spatial resolutions to reflect

the spatial and temporal variability of the hazard (Evans et al. 2008). However, hazard

mapping is traditionally conducted with a limited number of DRIs that are transported

through time and space during surveys. This practice requires data interpolation because

measurements likely fail to portray the temporal variability present (Koehler and

Volckens 2011; Lake et al. 2015), and in some cases temporal variability can be

incorrectly interpreted as spatial variability (Ludwig et al. 2017). Sensor networks have

the potential to avoid a major pitfall in hazard mapping by reducing errors due to data

sparsity (or “completeness”) that arise from the inability to measure a hazard at all

locations and times simultaneously (Koehler and Volckens 2011; Lake et al. 2015). In

addition to errors of completeness, DRIs are prone to other errors also, including poor

accuracy or precision, lack of sensitivity and biases from interferences (Koehler and

Volckens 2011). Using sensors to map hazards instead of DRIs will likely result in larger

errors of these types.

In this study, we report on the deployment of a sensor network in a heavy-vehicle

manufacturing facility, capable of measuring multiple agents of occupational interest

including PM, hazardous gases and noise simultaneously and in real time, for a study

period of 5 months. Although low-cost sensor networks have previously been used in the

49

general environment, our network is the first of which we are aware that has been

deployed in the industrial setting.

METHODS

Multi-hazard Monitor

We designed and constructed the 50 multi-hazard monitors and deployed 40 in this

network (Thomas et al. 2018). Briefly, each monitor was equipped with a dust sensor to

measure PM (GP2Y1010AU0F, SHARP Electronics, Osaka, Japan); an oxidizing gas (O3

+ NO2) sensor (OX-B431, Alphasense Ltd., Essex UK); a CO sensor (CO-B4,

Alphasense Ltd., Essex UK); a custom sound pressure level (SPL) sensor to measure

noise (Hallett et al. 2018); and a temperature and relative humidity sensor (AM2302,

Adafruit, New York, NY). A microcontroller (Seeeduino Cloud, Seeed Technology Co.,

Ltd., Guangdong, P.R.C) was programmed to read the electric signals from each sensor

every 2 seconds and then average the signals and wirelessly transmit the averaged data to

a central database approximately every 5 minutes.

Sensor Calibration

Due to constraints at the facility, individual calibration of each PM sensor in the network

was not feasible. Instead, the PM sensors used in the study were selected for inter-sensor

agreement in the laboratory, and then underwent field calibration to translate sensor

response to respirable PM concentration. Briefly, 100 sensors underwent a 6-point

laboratory calibration with dried salt particles. From these experiments, we selected 50

sensors with the most similar calibration slopes (all 50 sensors within ±14% of the

50

average slope of all 100 sensors) for the network. Three monitors were selected for a 2-

week field PM calibration procedure with a nephelometer (pDR-1000, Thermo Scientific,

Franklin, MA). The 3 monitors and pDR-1000 were deployed on the same I-beam in the

facility. The mean of the 3 zero-corrected PM sensor responses was correlated to the

aerosol concentration measured by the pDR-1000 using ordinary least squares linear

regression. The pDR-1000 measurements were corrected with a 6-hour gravimetric

respirable dust filter sample. The field calibration equation derived from this procedure

was then applied to all of the PM sensors in the network for the duration of the present

study.

For the CO and oxidizing gas sensors, we developed calibration curves in the laboratory

using a sample of 3 sensors of each type (Afshar-Mohajer et al. 2018). Briefly, we

exposed the sensors to concentrations of the target gases in a chamber and correlated

sensor response with a reference instrument. Although we developed calibration curves

for the OX-B421 sensors with both ozone (O3) and nitrogen dioxide (NO2) because the

sensor responds to both gases without discrimination, we applied the calibration slope for

O3. The calibration curves generated from a sample of 3 sensors of each type were then

applied to all sensors of that type in the monitor network for the duration of the present

study.

The noise sensor developed for this monitor network used a Microprocessor (Teensy 3.2,

open source) with an omnidirectional condenser microphone (CMA-4544PF-W, CUI

Inc., Tualatin, OR) (Hallett et al. 2018). Briefly, each of the sensors were calibrated by

51

playing “pink noise” with an acoustic generator (TalkBox, NTi Audio AG, Liechtenstein)

and an amplifier (Fender Musical Instruments Corp., Scottsdale, AZ) between 65 and 95

dB in 5 dB increments. Sensor response was compared to the collocated reference sound

level meter (XL2, NTi Audio AG, Liechtenstein) for an acceptance criterion of ±2 dB.

The results of this calibration procedure were applied for the duration of the present

study.

Monitor Network Deployment

The network was installed within an 806,400 square-foot (74,917 m2) area of a more than

2-million square-foot (185,806 m2) manufacturing facility that produces heavy vehicles

for construction and forestry. The monitors in our network were deployed in the facility

in a spatially-optimized pattern to capture maximum spatial variability and reduce

monitor redundancy (Berman et al. 2018). Briefly, researchers conducted seven mapping

events of approximately two hours in duration where particle number concentrations

were measured with a condensation particle counter (CPC; model 3007, TSI Inc.,

Shoreview, MN) and respirable mass concentrations were measured with an optical

particle counter (OPC; model PDM-1108, Grimm, Ainring, Germany) at 80-82 locations

on the manufacturing floor. Kriged hazard maps created from these seven field events

characterized the spatial variability and correlation structure of PM in the facility. A

methodology was applied to determine which of the 82 locations could be removed while

still maintaining accuracy and precision of the resulting hazard maps. The best hazard

map was considered one that prioritizes locations with high standard deviations (large

temporal variability at measurement locations) and high prediction precision (low kriging

52

variance). From this work, the optimal locations for monitors in our network on the

manufacturing floor were determined, reducing the number of monitor locations needed

to produce optimal hazard maps.

Forty monitors were deployed at 38 locations on regularly-spaced structural I-beams

throughout the manufacturing floor in locations closest to those optimal locations as was

practicable. Examples of instances where a monitor was not placed at the optimal

location included lack of a power outlet for the monitor or I-beam inaccessibility due to

obstruction by equipment or construction. An inventory of the manufacturing processes

surrounding was used to group monitors. The groups and the number of monitors in each

group were: machining (5), machining and welding (9), manual welding (11), manual

welding and robotic welding (5), staging (2), shot blasting (4), flame cutting (1), shot

blasting and laser cutting (3). One central location in the facility was chosen to collocate

3 monitors, allowing us to evaluate the precision of each type of sensor. At this location

we also performed the field calibration routine for the PM sensors.

Data Processing and Mapping

All data analysis was performed with MATLAB R2017a (Natick, MA). We identified

and removed database measurements from malfunctioning sensors, identified by

‘flatlined’ or abnormal signals. Five-minute data were averaged and all further

calculations were performed on 1-hr averages. Data were grouped according to

manufacturing processes occurring within a radius of ≤ 60 ft (18 m) of each monitor. We

created violin plots to examine the within- and between- group variability of hazards as

53

well as the distribution of hazard levels within groups. To construct hazard maps, 1-hour

means from each monitor were plotted and an inverse-distance weighting routine was

used to interpolate the hazard level at unmeasured locations. These maps were compiled

into videos that displayed space-time trends in hazard levels in the study area.

Network Precision

Precision of network hazard estimates from the 3 collocated monitors were examined in 2

ways. We plotted the difference between each individual monitor and the mean of the 3

monitors against the mean of the 3 monitors, which displays the range of hazard

estimates at a given concentration or SPL. We also plotted the second-order coefficient of

variation (V2) (Kvålseth 2017) against the mean of the 3 monitors for 5-minute and 1-

hour average network measurements. The coefficient of variation (V) is defined as:

𝑉 = 𝜎

𝜇 (1)

Where σ is the standard deviation and μ is the mean response of collocated monitors. The

second-order coefficient of variation was calculated as:

𝑉2 = (𝑉2

1+𝑉2)

12⁄

(2)

V2 has bounds from 0 to 1 and approximates the coefficient of variation up to a value of

about 0.45 (where V = 0.50). Beyond a value of 0.45, V and V2 increasingly diverge. We

54

plotted V2 against the mean to display variation of sensor measurements in a format

comparable to coefficient of variation able to accommodate the high levels of variability

observed at concentrations near zero.

Network Accuracy

We compared hazard measurements from each node of the network to measurements

from direct-reading reference instruments in both August and December to assess the

accuracy of the monitor network. The comparison consisted of collocating measurements

with each monitor and reference instruments for 1-minute on the first occasion and for 5-

minutes on the second occasion. We bypassed the database for this procedure in order to

collect 2-second data directly from each monitor via serial connection with a computer.

The reference instruments were as follows: respirable PM, personal DataRAM 1500

configured for respirable dust sampling (pDR-1500, Thermo Scientific, Franklin, MA);

CO, Q-Trak 7575 (TSI Inc., Shoreview MN); Personal Ozone Monitor (‘POM,’

2BTechnologies, Boulder, CO); and noise, model XL2 (NTi Audio AG, Liechtenstein).

For each monitor, we computed the mean signal from each sensor and converted the

output to concentration for the PM, CO and oxidizing gas sensors using calibration

protocols described above (the noise sensor output was dBA). Bias, B, was calculated for

each hazard with respect to a reference instrument for each monitor according to:

𝐵 =𝜇

𝐶𝑇− 1 (3)

55

where μ is the mean hazard level measured by the low-cost sensor and CT is the “true

concentration” of the hazard level measured by the reference instrument (NIOSH 2012).

Percent biases were plotted against the reference instrument concentration or SPL to

evaluate accuracy across the range of hazard levels observed during the validation

campaigns and evaluated against the NIOSH accuracy criterion of bias within ±10%

(NIOSH 2012).

RESULTS

Temporal Variability

The multi-hazard monitor network was continuously deployed for 5 months (August 4,

2017 – January 8, 2018) and recorded over 1.46 million measurements of PM, CO,

oxidizing gases and noise to the database. Over this period of time, the network captured

the diurnal and weekly patterns of all hazards (Figure 3.1; gray lines represent individual

monitors and the black line shows the mean over all monitors). The temporal variability

observed for each hazard was consistent with manufacturing activities in the facility, such

as daily peaks, decreases during overnight periods and weekend low concentrations. The

mean hazards were also correlated with one another (Pearson’s correlation coefficient

0.56 ≤ r ≤ 0.84; Appendix 3.1), despite differences in their accumulation, distribution,

and dissipation. The mean daily maximum 1-hr PM concentrations ± standard deviation

(SD) recorded across all monitors was 0.52 ± 0.1 mg/m3 on typical production days, with

individual monitors recording measurements up to 4.3 mg/m3. The mean weekend low ±

SD was 0.17 ± 0.09 mg/m3. The mean daily maximum 1-hr CO concentrations ± SD

observed across all monitors was 7 ± 2 ppm, with some network nodes occasionally

56

reaching the 12-ppm ceiling of the sensor in this network, and mean concentrations below

1 ppm on weekends. The mean daily maximum 1-hr oxidizing gas concentrations ±

standard deviation observed across all monitors in the network ranged from 125 ± 27 ppb

with individual monitors recording measurements of up to 560 ppb O3 + NO2 and mean

weekend lows ± SD equal to 17 ± 19 ppb. The mean daily maximum 1-hr noise SPLs ±

SD recorded by all monitors in the network was 83 ± 1 dBA with individual monitors

detecting up to 93 dBA and lows on the weekend of 73 ± 2 dBA. The mean daily

maximums of PM and CO were normally distributed, but oxidizing gases and noise were

not. We observed changes in the baseline for each sensor type, characterized as the mean

sensor response during the weekends over the course of the deployment when hazard

levels were consistently low. Differences in the mean hazard level in the weekend

following the first normal production week after deployment and the weekend following

the last normal production week before the end of the study period were 0.09 mg/m3 for

PM, 0.3 ppm for CO, 3 ppb for oxidizing gases and 2 dBA for noise. We also observed a

differential loss of precision among the different types of sensors over the course of

deployment. There was an increase in SD among all sensors of the same type between the

same weekends equal to 0.27 mg/m3 for PM, 0.2 ppm for CO, 19 ppb for oxidizing gases,

and 0.3 dBA for noise.

Intra- and Inter-Group Variability

The variability of PM, CO, oxidizing gases and noise within and between the groups of

work processes for typical production periods are displayed in Figure 3.2. For PM, there

were differences in the median PM concentration between groups as high as 0.33 mg/m3

57

for the difference between flame cutting (FC) and staging (S) areas. There was also a

difference in the variability of PM concentrations within the groups of work processes.

For example, the interquartile range (IQR) of PM concentrations in the staging area was

0.16 mg/m3, compared to 0.54 mg/m3 in the shot blasting (SB) area. We also observed

PM concentrations that were bimodal in some work areas, such as machining (M) and

manual welding and robotic welding (MW&RW). Median CO concentrations during

typical production periods were varied the most between areas with manual welding

(MW) and manual welding and robotic welding (MW&RW) (6 ppm) and shot blasting

and laser cutting (SB&LC) (4 ppm). The lowest CO variability was observed in the flame

cutting (FC) area (IQR = 2 ppm) compared to the manual welding and robotic welding

(MW&RW) areas (IQR = 3 ppm). The groups of monitors with the greatest oxidizing gas

concentration was machining and welding (M&W) (117 ppb) and the lowest oxidizing

gas concentration was shot blasting and laser cutting (SB&LC) (72 ppb). With oxidizing

gases as well, we observed variability in the distributions of concentrations measured by

the network. The shot blasting (SB) area had an IQR equal to 51 ppb, whereas in the

flame cutting (FC) area the IQR was equal to 32 ppb. Noise variability throughout the

facility during typical production times was low, with the largest median SPL difference

occurring between the staging areas (S) (79 dBA) and machining and welding areas

(M&W) (83 dBA). The variability of noise within groups was more similar than other

hazards, with IQRs of SPLs in all manufacturing areas ranging from 1 to 3 dBA.

Space-time Variability

58

Examples of hazard maps created for each hazard with data from the network during

production periods are displayed in Figure 3.3. There was a daily pattern observed for

each hazard’s accumulation, peak and dissipation. For example, PM concentrations

originate, spread from and remain highest in areas with machining and welding as

primary work processes as seen in the upper center of the hazard maps. Machining in this

facility occurs on a large scale with substantial amount of cutting oil contributing to the

PM. For CO, concentrations are highest first, then spread, from areas in the lower left

quadrant of the facility where laser cutting and shot blasting are performed, cutting is a

combustion process and may produce CO. Oxidizing gases originate from areas of the

facility where many manual welding stations are present (in the center and upper right

quadrant of the hazard maps), mix throughout the facility and dissipate quickly during

breaks and at the end of production shifts. Welding arc produces both NO2 and O3. Noise

contrasts with these heterogeneous space-time patterns of PM, CO and oxidizing gases.

Noise in the facility increased uniformly throughout the facility from mean overnight

lows of 79 dBA to a mean of 83 dBA throughout production times. In contrast to the

other hazards under study where specific manufacturing processes contribute to hazard

levels, noise is produced everywhere and by nearly all processes in the facility. Noise is a

physical hazard that does not disperse from a source the same ways that particulate or

gaseous hazards do, leading to the more homogenous SPLs observed in this study. The

network has limited ability to capture impact or impulse noise (sudden, brief SPLs

exceeding 140 dB) because of the 5-minute average SPL recorded to the database,

however examination of 5-minute average SPL data (not shown here) does show brief

spatially restricted increases in SPL.

59

Network Precision

Among collocated monitors, we observed 95% of measurements were within 0.23 mg/m3

for PM, 0.4 ppm for CO, 7 ppb for oxidizing gases, and 1 dBA for noise (Figure 3.4). We

observed slightly smaller differences between the 3 monitors at lower concentrations of

PM, CO and oxidizing gases. For noise, the difference in SPL measurements among the 3

monitors was similar across the range of SPL observed. The second-order coefficient of

variation for all hazards is displayed in Figure 3.5. The median V2 of 1-hr average

measurements for PM, CO, oxidizing gas and noise of the 3 collocated monitors was

0.29, 0.02, 0.02, and 0.004, respectively. For all hazards generally over the duration of

the study period, the V2 did not change beyond daily and weekly patterns associated with

hazard levels. The collocated PM sensors displayed the greatest variability of the 4

hazards across the range of observed concentrations, and at concentrations of 0.2 and 0.4

mg/m3, the V2 was approximately equal to 0.30 and 0.18, respectively. In general, there

was a smaller range of V2 at a given concentration for 1-hr average measurements

compared to the 5-minute data.

Network Accuracy

The bias of the monitors collocated with field reference instruments is shown in Figure

3.6 and varied among the hazards under study. The magnitude of the median percent bias

between network monitors and field reference instruments were equal to 41%, 7%, 36%

and 1%, for PM, CO, oxidizing gases and noise respectively. For PM, we observed the

magnitude of the percent bias decrease rapidly from a high of 524% with increasing

60

concentrations, with 12% of the measurements meeting the NIOSH bias criteria of

percent bias within 10%. For concentrations of CO up to 12 ppm, the magnitude of the

bias for all but 1 collocated measurement was within 25%, and 58% of validation

measurements met the NIOSH bias criterion. For concentrations greater than 12 ppm CO,

the observed bias was greater because the concentration was beyond the CO

concentration ceiling for these sensors as operated in our network. We observed the

magnitude of the bias associated with oxidizing gas concentrations from the OX-B421

sensor and the O3 reference instrument ranging from 0.33% to 156%. Some of the bias is

explained by the fact that the OX-B431 sensor responds to both NO2 and O3 with CO2 as

a major interferent (Lewis et al. 2016), compared to the POM which is highly specific to

O3. We experienced a malfunction of the POM preventing O3 bias calculations for

December collocation measurements. In this reduced number of collocated

measurements, 13% met the NIOSH bias criterion. The noise sensor outperformed the

PM, CO and oxidizing gas sensors with respect to the magnitude of the bias, which was

between 0.04-4.80% over the range of SPLs observed in the collocated measurements,

and all (100%) measurements met the NIOSH bias criterion.

DISCUSSION

To our knowledge, this is the first multi-hazard monitor network constructed with low-

cost sensors deployed in an industrial setting. This study demonstrates the ability of

sensor networks to capture the temporal and spatial patterns of occupational hazards that

traditional industrial hygiene approaches would not. The hazard maps that were produced

with data from the network offer insight into the sources, areas of high concentration, and

61

distribution of hazards and could be used to evaluate if control strategies are effective,

offering another advantage over traditional industrial hygiene approaches.

In contrast to networks deployed in the ambient environment, this multi-hazard network

was deployed in a setting where the distances between monitors were small (less than 135

ft with a mean distance to nearest monitor ± SD = 92 ± 26 ft), and the concentration of

pollutants was high. PM concentrations, for example, in this facility were high enough to

foul some sensors and cause signal baseline drift after relatively short periods of time

(Thomas et al. 2018). Another challenge with low-cost sensors that may be exacerbated

by high concentration environments is that the variability in sensor measurements may be

greater than that of the mean levels of the pollutant under study (Lewis and Edwards

2016). Here, we demonstrate that is the case with PM, where 28% of the maximum

differences in 1-hr average PM concentrations among 3 collocated monitors were greater

than the mean of the concentrations observed for 3 collocated monitors (Figure 3.4).

A major challenge in mapping occupational hazards is that temporal and spatial

variability both contribute to measurement uncertainty (Koehler and Volckens 2011). As

we have demonstrated here, sensor and monitor networks have the potential to overcome

this challenge and provide highly temporally and spatially resolved measurements of

pollutants – our network recorded PM, CO, oxidizing gas and noise levels at 5-minute

time intervals and was spatially optimized to reduce uncertainty of hazard measurements.

A major pitfall of hazard mapping is a lack of data completeness which may lead to

incorrect conclusions, underperforming control interventions or wasted resources for

62

surveillance and measurement (Koehler and Volckens 2011). Such consequences can be

ameliorated by using sensor or monitor networks which address lack of completeness due

the nature of a network’s individual nodes being distributed throughout the facility and

ability to take measurements simultaneously.

In this study, diurnal and weekly patterns of occupational hazards became apparent after

several weeks of deployment, and the full 5 months of data were not needed to establish

the repeating patterns of hazards in this facility. In work environments with relatively

constant or regular levels of production, a guideline of a 1-month network deployment

could be reasonably used to establish space-time patterns of occupational hazards. We

caution against attempting to infer such patterns during times of reduced production, such

as holidays or other shutdowns. In our long-term network deployment examples of these

periods of time are clearly visible in Figure 3.1 including the August manufacturing

shutdown when the network was initially deployed, the American Thanksgiving holiday

in late November and the Christmas/New Year holiday shutdown in late December and

early January. Surprisingly, we did not observe a strong seasonal influence on hazard

levels, which might reasonably be expected due to changes in heating, ventilation and air

conditioning (HVAC) practices in the facility. An advantage of the long-term deployment

of a sensor or monitor network is that this kind of long-term space-time variability can be

explicitly characterized in a way that intermittent measurements or measurements at a

limited number of locations cannot. Another advantage offered by our network was the

ability to observe the spatial variation of hazards and associate their concentrations with

specific manufacturing processes.

63

The use of low-cost sensors in wireless networks poses many challenges including the

need for thorough laboratory and field calibration, sensor baseline drift over time, and

overall lower data quality (Lewis and Edwards 2016; Lewis et al. 2016; Masson et al.

2015; Piedrahita et al. 2014; Snyder et al. 2013; Xiong and Compton 2014). Another

challenge, for gas sensors in particular, is sensor specificity to the target gas, which

produces erroneous response from interfering gases (Masson et al. 2015; Spinelle et al.

2015a). In this network, the oxidizing gas sensor signal is a summation of response to

both O3 and NO2, and is unable to discriminate between the two gases (Afshar-Mohajer

et al. 2018; Hossain et al. 2016). The oxidizing gas sensor’s non-specific response makes

accurate estimation of ozone quite challenging and a comparison to the PEL difficult at

best and misleading at worst. For example, a similar response from the OX-B431 sensor

to NO2 at the mean to maximum levels of response observed in this study in the absence

of O3 would be associated with approximately 0.1-0.5 ppm of NO2, well below the PEL

(ceiling) of 5 ppm NO2. Future multi-hazard monitor networks may attempt to improve

O3 concentration estimates by employing pared electrochemical sensors, one oxidizing

gas sensor (O3 + NO2), like was used in our network, and one sensor specific to NO2

(Hossain et al. 2016).

Sensor calibration in sensor networks with a large number of sensor nodes is a key

consideration in their deployment. Users have three main calibration options: 1) users can

apply the manufacturer’s calibration constants for slope and intercept 2) create a

calibration curve specific to each sensor in the field or in the laboratory, or 3) develop a

64

calibration curve that can be generalized to all sensors in a given lot of sensors in the

field or laboratory. The advantage of applying a common calibration curve to all sensors

in a network includes a simplification of data processing and translation of sensor signal

to concentration and the avoidance of calibrating each sensor in the sensor network. On

the other hand, using a common calibration curve based on a sample of sensors

introduces a source of measurement error in a sensor network because of variability in the

response of sensors of the same type. We suggest calibrating a sample of sensors set up in

their intended configuration and evaluating if the variability in the calibration slope of the

sample of sensors exceeds the tolerance of acceptable measurement error for a given

application. This strategy does however require some prior knowledge about the

concentration of target gas in the environment of interest and is the subject of future

work.

Unfortunately, none of these 3 sensor calibration strategies solve or take into account that

the calibration slope of some types of sensors change over time (Afshar-Mohajer et al.

2018), calibration relationships may only hold for specific locations for a limited period

of time (Lewis and Edwards 2016), or calibration may differ substantially in the

laboratory versus the field (Piedrahita et al. 2014). A limitation of this study is that the

same calibration slopes were used throughout this study for the PM, CO and OX sensors,

which likely introduced increasing measurement error over time. Future work should

consider how in-field assessment can be used to update the calibration of low-cost

sensors, for example with the collocation of a higher-quality field reference instrument or

the use of calibration gases.

65

Concentrations of indoor contaminants are highly variable, autocorrelated in time and

space, and related to occupant activities, which complicates statistical procedures and

likely leads to an underestimation of variance (Francis et al. 1989; Høst et al. 1995;

Kolovos et al. 2010; Luoma and Batterman 2000; Symanski and Rappaport 1994).

Although a variety of statistical techniques have been applied to address these issues of

non-independence and autocorrelation, in this preliminary data analysis we have

interpolated hazard concentrations between measured locations, using inverse-distance

weighting, a technique that ignores spatial and temporal autocorrelation in the mapped

measurements (Koehler and Peters 2013). In addition, using inverse-distance weighting

does not permit a relationship of the hazard with respect to distance other than the inverse

of distance raised to some power, the calculation of standard errors for hazard estimates,

or the incorporation of other variables into the hazard prediction (e.g. manufacturing

processes).

In future work we will apply geostatistical approaches such as kriging to the dataset

collected with the sensor network. We will first investigate if there is sufficient data for

such approaches by evaluating the semivariogram of each hazard dataset to characterize

spatial dependence. Subsequently, we will be able to examine the correlation structure of

the hazards measured with this network, characterize the statistical variability in mapped

hazard levels and investigate influences of variables other than distance that are

associated with hazard levels. Geostatistical approaches, given there is sufficient data for

their appropriate application, provide more information compared to non-statistical

66

approaches such as inverse-distance weighting (e.g. the error or confidence associated

with hazard predictions), and offer the opportunity to improve the accuracy of hazard

predictions by the incorporation of other predictor variables or non-inverse-distance

weighted relationships.

CONCLUSIONS

Here we demonstrated the ability of a spatially dense (maximum distance to the nearest

monitor equal to 135 ft [41 m]) sensor network to collect information over 5 months on

multiple occupational hazards at a time interval of 5 minutes. Examination of network

data provided insight into the daily, weekly, and seasonal patterns and the spatial

distribution of hazards in the facility including hotspot identification that wouldn’t be

possible with traditional industrial hygiene approaches. It also allowed us to examine the

manufacturing processes associated with higher levels of the various hazards in the

network. Despite these successes, serious challenges with sensor accuracy, precision,

stability over time, and cross-sensitivity to non-target species persists. In campaigns to

verify the accuracy of the network, we observed a range of bias with respect to high

quality direct reading instruments depending on the hazard and the concentration/level,

with median biases ranging from 1% for noise to 41% for PM. Within a set of 3

collocated monitors in the network, we observed a range in precision by hazard and

absolute differences between monitors that tended to be greater at higher hazard levels

and relative differences between monitors that were higher at low hazard concentrations.

These lessons learned in this study, as well as the account of our experience are

generalizable to others who wish deploy sensor networks in occupational environments.

67

Future work will investigate the feasibility of using a sensor network to quantitatively

estimate personal exposure in the occupational environment.

68

FIGURES

Figure 3.1. Time series of 1-hr average hazard concentrations/intensities measured by the multi-hazard monitor network.

Grey shaded lines are measurements from each individual monitor and black lines display the mean of all monitors.

69

Figure 3.2 (next page). Distribution of hazard level by manufacturing process.

Typical 1-hr measurements during shift 1 from August 14 – Dec 22, 2017 (time

excludes weekends, holidays and shutdown periods). Monitors are grouped by

major work processes occurring within an 80 x 120 ft area surrounding each

monitor. Manufacturing process abbreviations: machining (M), machining and

welding (M&W), manual welding and robotic welding (MW&RW), staging (S), shot

blasting (SB), flame cutting (FC), shot blasting and laser cutting (SB&LC).

70

71

Figure 3.3. 1-hr Hazard Maps on the morning of August 17, 2017; the day of August validation routine. For each hazard, the

map on the left shows concentrations/intensities before the shift starts and the plot on the right shows concentrations during

work operations. Circles represent locations of network nodes.

72

Figure 3.4. Precision of 1-hr average measurements among collocated monitors. Each color represents a different monitor.

73

Figure 3.5. Second-order coefficient of variation (V2) plotted against mean measurement of three collocated sensors for 5-min

(grey) and 1-hr (black) averaging time.

74

Figure 3.6. Sensor measurement accuracy is shown as %Bias against the concentration/intensity measured by the reference

instrument. Circles from August 17, 2017 with 1-min collocated measurements and squares from December 21 and 22, 2017

with 5-min collocated measurements.

75

REFERENCES



ozone, nitrogen dioxide, and carbon monoxide. Journal of Occupational and

Environmental Hygiene 15:87-98.

Anderson JO, Thundiyil JG, Stolbach A. 2012. Clearing the air: A review of the effects

of particulate matter air pollution on human health. Journal of Medical Toxicology

8:166-175.

Berman JD, Peters TM, Koehler KA. 2018. Optimizing a sensor network with data from

hazard mapping demonstrated in a heavy-vehicle manufacturing facility. Ann Work

Expo Health 62:547-558.

Bornholdt J, Dybdahl M, Vogel U, Hansen M, Loft S, Wallin H. 2002. Inhalation of

ozone induces DNA strand breaks and inflammation in mice. Mutation

Research/Genetic Toxicology and Environmental Mutagenesis 520:63-72.

Dockery DW. 1993. Epidemiologic study design for investigating respiratory health

effects of complex air pollution mixtures. Environmental health perspectives 101

Suppl 4:187-191.

English PB, Olmedo L, Bejarano E, Lugo H, Murillo E, Seto E, et al. 2017. The imperial

county community air monitoring network: A model for community-based

environmental monitoring for public health action. Environmental health perspectives

125.

Evans DE, Heitbrink WA, Slavin TJ, Peters TM. 2008. Ultrafine and respirable particles

in an automotive grey iron foundry. The Annals of Occupational Hygiene 52:9-21.

76

Francis M, Selvin S, Spear R, Rappaport S. 1989. The effect of autocorrelation on the

estimation of workers' daily exposures. The American Industrial Hygiene Association

Journal 50:37-43.

Gao M, Cao J, Seto E. 2015. A distributed network of low-cost continuous reading

sensors to measure spatiotemporal variations of pm2. 5 in xi'an, china. Environmental

pollution 199:56-65.

Hallett L, Tatum M, Thomas G, Sousan S, Koehler K, Peters T. 2018. An inexpensive

sensor for noise. J Occup Environ Hyg 15:448-454.

Harper M. 2004. Assessing workplace chemical exposures: The role of exposure

monitoring. Journal of Environmental Monitoring 6:404-412.

Hasenfratz D, Saukh O, Walser C, Hueglin C, Fierz M, Arn T, et al. 2015. Deriving high-

resolution urban air pollution maps using mobile sensor nodes, pervasive and mobile

computing, v. 16 n.

Heidemann J, Bulusu N. Using geospatial information in sensor networks. In:

Proceedings of the Proc CSTB workshop on Intersection of Geospatial Information

and Information Technology, 2001.

Heimann I, Bright VB, McLeod MW, Mead MI, Popoola OAM, Stewart GB, et al. 2015.

Source attribution of air pollution by spatial scale separation using high spatial

density networks of low cost air quality sensors. Atmospheric Environment 113:10-

19.

Heitbrink WA, Evans DE, Peters TM, Slavin TJ. 2007. Characterization and mapping of

very fine particles in an engine machining and assembly facility. Journal of

occupational and environmental hygiene 4:341-351.

77



Høst G, Omre H, Switzer P. 1995. Spatial interpolation errors for monitoring data.

Journal of the American Statistical Association 90:853-861.

Ikram J, Tahir A, Kazmi H, Khan Z, Javed R, Masood U. 2012. View: Implementing low

cost air quality monitoring solution for urban areas. Environmental Systems Research

1:1.

Jiang Q, Kresin F, Bregt AK, Kooistra L, Pareschi E, van Putten E, et al. 2016. Citizen

sensing for improved urban environmental monitoring. Journal of Sensors 2016.

Jiao W, Hagler G, Williams R, Sharpe R, Brown R, Garver D, et al. 2016. Community air

sensor network (cairsense) project: Evaluation of low-cost sensor performance in a

suburban environment in the southeastern united states. Atmos Meas Tech 9:5281-

5292.

Jovašević-Stojanović M, Bartonova A, Topalović D, Lazović I, Pokrić B, Ristovski Z.

2015. On the use of small and cheaper sensors and devices for indicative citizen-

based monitoring of respirable particulate matter. Environmental Pollution 206:696-

704.

Kampa M, Castanas E. 2008. Human health effects of air pollution. Environmental




840.

78

Koehler KA, Peters TM. 2013. Influence of analysis methods on interpretation of hazard

maps. The Annals of Occupational Hygiene 57:558-570.

Kolovos A, Skupin A, Jerrett M, Christakos G. 2010. Multi-perspective analysis and

spatiotemporal mapping of air pollution monitoring data. Environmental science &

technology 44:6738-6744.



2610.




Kvålseth TO. 2017. Coefficient of variation: The second-order alternative. Journal of

Applied Statistics 44:402-415.

Lake K, Zhu J, Wang H, Volckens J, Koehler KA. 2015. Effects of data sparsity and

spatiotemporal variability on hazard maps of workplace noise. Journal of


Lee D-D, Lee D-S. 2001. Environmental gas sensors. IEEE Sensors Journal 1:214-224.





Lippmann M. 1989. Health effects of ozone a critical review. JAPCA 39:672-695.

79

Liu S, Hammond SK. 2010. Mapping particulate matter at the body weld department in

an automobile assembly plant. Journal of occupational and environmental hygiene

7:593-604.

Ludwig G, Chu T, Zhu J, Wang H, Koehler K. 2017. Static and roving sensor data fusion

for spatio-temporal hazard mapping with application to occupational exposure

assessment. The Annals of Applied Statistics 11:139-160.

Luoma M, Batterman SA. 2000. Autocorrelation and variability of indoor air quality

measurements. AIHAJ-American Industrial Hygiene Association 61:658-668.






Moltchanov S, Levy I, Etzion Y, Lerner U, Broday DM, Fishbain B. 2015. On the

feasibility of measuring urban air pollution by wireless distributed sensor networks.

Science of The Total Environment 502:537-547.




NIOSH 2017. NIOSH manual of analytical methods (NMAM).National Institute for

Occupational Health and Safety.

80




OSHA 1974. OSHA noise regulations (standards-29 cfr): Occupational noise exposure.-

1910.95.

OSHA 1993. OSHA safety and health standards. 29 cfr 1910.1000. Table z-1. 1:444.

Ott DK, Kumar N, Peters TM. 2008. Passive sampling to capture spatial variability in pm

10–2.5. Atmospheric Environment 42:746-756.

Park JY, Ramachandran G, Raynor PC, Olson Jr GM. 2010. Determination of particle

concentration rankings by spatial mapping of particle surface area, number, and mass

concentrations in a restaurant and a die casting plant. Journal of occupational and

environmental hygiene 7:466-476.

Passchier-Vermeer W, Passchier WF. 2000. Noise exposure and public health.

Environmental health perspectives 108 Suppl 1:123-131.




Peters TM, Anthony TR, Taylor C, Altmaier R, Anderson K, O'Shaughnessy PT. 2012.

Distribution of particle and gas concentrations in swine gestation confined animal

feeding operations. The Annals of Occupational Hygiene 56:1080-1090.




81

Pope CA, Dockery DW, Schwartz J. 1995. Review of epidemiological evidence of health

effects of particulate air pollution. Inhalation toxicology 7:1-18.

Pope III CA, Dockery DW. 2006. Health effects of fine particulate air pollution: Lines

that connect. Journal of the Air & Waste Management Association 56:709-742.

Raub JA, Mathieu-Nolf M, Hampson NB, Thom SR. 2000. Carbon monoxide

poisoning—a public health perspective. Toxicology 145:1-14.




Sousan S, Koehler K, Thomas G, Park JH, Hillman M, Halterman A, et al. 2016. Inter-

comparison of low-cost sensors for measuring the mass concentration of occupational

aerosols. Aerosol Science and Technology 50:462-473.

Spinelle L, Gerboles M, Aleixandre M. 2015. Performance evaluation of amperometric



Symanski E, Rappaport SM. 1994. An investigation of the dependence of exposure

variability on the interval between measurements. The Annals of Occupational

Hygiene 38:361-372.



Vosburgh DJH, Boysen DA, Oleson JJ, Peters TM. 2011. Airborne nanoparticle

concentrations in the manufacturing of polytetrafluoroethylene (ptfe) apparel. Journal

of occupational and environmental hygiene 8:139-146.

82

Weschler CJ. 2006. Ozone's impact on public health: Contributions from indoor

exposures to ozone and products of ozone-initiated chemistry. Environmental health

perspectives:1489-1496.

Xiong L, Compton RG. 2014. Amperometric gas detection: A review.

IntJElectrochemSci 9:7152-7181.

83

CHAPTER FOUR

Manuscript 3

(In preparation for submission for publication in Environmental Science & Technology)

Estimating Personal Exposure with a Multi-Hazard Sensor

Network

Christopher Zuidema, Larissa V Stebounova, Sinan Sousan, Alyson Gray, Oliver Stroh,

Geb Thomas, Thomas Peters and Kirsten Koehler

84

ABSTRACT

Occupational exposure assessment is almost exclusively accomplished with personal

sampling. However, personal sampling can be burdensome and suffers from low sample

sizes, resulting in inadequately characterized workplace exposures. Sensor networks offer

the opportunity to measure occupational hazards with a high degree of space-time

resolution. Here, we demonstrate an approach to estimate personal exposure to particulate

matter, carbon monoxide, ozone, and noise using hazard data from a sensor network. We

simulated stationary and mobile employees that work at the study site, a heavy-vehicle

manufacturing facility. Network-derived exposure estimates compared favorably to

measurements taken with a suite of reference direct-reading instruments (DRIs) deployed

to mimic personal sampling but varied by hazard and type of employee. The median

magnitude of the percent bias between network-derived exposure estimates and DRI

measurements for stationary employees was 32% for PM, 23% for CO, 141% for O3, and

2% for noise; and for mobile employees was 36% for PM, 18% for CO, 119% for O3, and

3% for noise. Correlation between network-derived exposure estimates and DRI

measurements ranged from 0.39 (noise for mobile employees) to 0.75 (noise for

stationary employees). Despite the error observed estimating personal exposure to

occupational hazards it holds promise as an additional tool to be used with traditional

personal sampling due to the ability to frequently and easily collect exposure information

on many employees.

85

INTRODUCTION

Occupational environments, especially heavy industry, often have complex hazardous

exposures resulting from manufacturing processes including welding, cutting, grinding,

machining, and abrasive blasting. Exposures resulting from these processes include

particulate matter (PM); gases such carbon monoxide (CO), oxides of nitrogen (NOx),

and ozone (O3); metals including lead, nickel, zinc, iron oxides, copper, cadmium and

chromium; physical hazards such as noise, heat, electrical and vibration; and radiological

including visible and ultraviolet frequencies of light (Sferlazza and Beckett 1991). To

assess compliance with occupational exposure limits to workplace hazards, employers

perform exposure monitoring, typically by personal sampling on individuals suspected to

have high exposure (Rappaport and Kupper 2008). However, personal sampling can have

drawbacks such as high expense and burden to employees and generally suffers from a

low number of samples taken (Rappaport 1984). In most cases, fewer than six samples at

an industrial facility are used to judge if employees may be over-exposed or workplaces

are in compliance with regulatory permissible exposure limits (Roick et al. 1991), and

many rely on just one measurement (Tornero‐Velez et al. 1997). This situation results in

inadequately characterized workplace exposures and occupational risks that may be

higher than compliance testing indicates (Rappaport 1984).

To ameliorate this problem, the National Institute for Occupational Safety and Health

(NIOSH) has called for “comprehensive exposure assessment,” where risks from all

hazards for all days and all workers are considered (Ramachandran 2008). Furthermore,

cost-efficient occupational exposure assessment, where both economics and statistical

86

efficiency (e.g. sample size and measurement error) are considered, is also needed

(Rezagholi and Mathiassen 2010). Low-cost sensors could potentially fill this need and

have recently attracted the attention of environmental health scientists seeking to measure

air pollution with a high degree of temporal and spatial resolution (Kumar et al. 2015;

Lewis et al. 2016; Masson et al. 2015; Piedrahita et al. 2014; Snyder et al. 2013).

Advances in open software toolkits and microprocessor platforms have facilitated the

development of customized wireless sensor networks, and there is a growing number of

examples in the literature (English et al. 2017; Gao et al. 2015; Hasenfratz et al. 2015;

Heimann et al. 2015; Ikram et al. 2012; Jiang et al. 2016; Jiao et al. 2016; Kumar et al.

2011; Mead et al. 2013; Moltchanov et al. 2015). Data from sensor networks can be used

to create hazard maps (Evans et al. 2008; Heitbrink et al. 2007; Liu and Hammond 2010;

O'Brien 2003; Ott et al. 2008; Park et al. 2010; Peters et al. 2006; Peters et al. 2012;

Vosburgh et al. 2011), which visually communicate risk (Koehler and Volckens 2011),

identify hazard sources (Evans et al. 2008; O'Brien 2003), characterize the distribution of

hazards in a facility or the environment (Evans et al. 2008; Ott et al. 2008; Peters et al.

2006), and inform hazard control strategies (O'Brien 2003).

We have previously developed a multi-hazard sensor network constructed with low-cost

sensors for PM, CO, oxidizing gases (O3 + NO2) and noise (Thomas et al. 2018). An

industrial hygienist identified hazards at the study site, a heavy-vehicle manufacturing

facility, and those of greatest occupational health importance were chosen for inclusion

our sensor network. PM has well-characterized associations with cardiopulmonary and

respiratory diseases, lung cancer, inflammation, oxidative stress, pulmonary infection,

87

and lung function (Anderson et al. 2012; Dockery 1993; Pope et al. 1995; Pope III and

Dockery 2006). The Permissible Exposure Limit (PEL) for respirable PM is 5 mg/m3

(OSHA 1993). The health effects of CO less than or equal to the PEL, which is equal to

50 ppm (OSHA 1993), include headache, dizziness, weakness, nausea and confusion

(Raub et al. 2000). The inhalation of O3 causes inflammation, reduced lung function,

DNA damage and increased symptoms and development of asthma (Bornholdt et al.

2002; Kampa and Castanas 2008; Lippmann 1989; Weschler 2006). The PEL for O3 is

100 ppb (OSHA 1993). Occupational noise exposure induces hearing impairment,

hypertension and annoyance (Passchier-Vermeer and Passchier 2000). Additionally,

there is limited evidence that noise in the workplace is associated with biochemical and

immune effects, and impacts absentee rate and performance (Passchier-Vermeer and

Passchier 2000). The permissible exposure level to noise for an 8-hr work period is 90

dBA (OSHA 1974).

In previous work, we have described the long-term deployment of our multi-hazard

sensor network capable of mapping PM, CO, oxidizing gases and noise at the study site

with a high degree of space-time resolution (Zuidema et al. in revision). In the current

study we demonstrate that hazard mapping data from a sensor network, when combined

with an individual’s location information, can be used to quantitatively estimate personal

exposure to multiple occupational hazards simultaneously. We compare the network-

derived hazard estimates to personal measurements collected from high-quality, portable,

direct-reading instruments that are used in traditional industrial hygiene practice.

88

METHODS

Sensor Network

We designed and constructed multi-hazard monitors, the sensors for which are

summarized in Table 4.1. Each monitor, or “node” of the sensor network was equipped

with sensors to measure PM (GP2Y1010AU0F, Sharp Electronics, Osaka, Japan);

oxidizing gases (OX-B431, Alphasense Ltd., Essex UK; responsive to both O3 and NO2);

CO (CO-B4, Alphasense Ltd., Essex UK); sound pressure level (SPL) (Hallett et al.

2018); and temperature and relative humidity (AM2302, Adafruit, New York, NY)

(Thomas et al. 2018). The 40-node network was installed for approximately 8 months

within 74,900 m2 (806,400 ft2) of a +185,800 m2 (+2-million ft2) manufacturing facility

that produces heavy vehicles for construction and forestry. The nodes of the network

were deployed in a spatially optimized pattern to capture maximum spatial variability

(Berman et al. 2018), and measurements from each monitor were transmitted wirelessly

to a central database approximately every five minutes, permitting the hazard variability

to be characterized with a high degree of spatial and temporal resolution. We have

previously reported on the spatial and temporal variability of hazards, sensor precision,

and measurement accuracy in the facility (Zuidema et al. in revision).

Worker Simulation and Reference DRIs

On five occasions in August 2017, December 2017, and March 2018 we simulated two

types of workers. The first type of worker was one that remained in a relatively small

geographic space (within an area of smaller than 12 x 18 m) to perform their work duties,

such as a welder or machine operator. For this type of simulated worker, direct-reading

89

instruments (DRIs) were deployed at an employee workstation for the duration of the

simulated work shift and were not moved. Hereafter we refer to this simulated employee

type as the “stationary” routine. The second type of worker was one that was highly

mobile and traveled throughout the facility at a walking or slow biking/driving pace, such

as supervisors, mechanics, employees that move small parts between workstations and

maintenance workers. For the second type of simulated worker, DRIs were worn by study

staff, as a worker would for traditional personal sampling. Hereafter we refer to this

simulated employee type as the “mobile” routine. For the mobile sampling routines, study

staff kept a detailed log of their position as they moved throughout the facility according

to an established coordinate system, marked by regularly spaced structural I-beams. We

simulated a total of 22 work shifts (19 stationary and 5 mobile) during times of typical

production (weekdays, 6:00-16:00). The DRIs were as follows: respirable PM, personal

DataRAM 1500 configured for respirable dust sampling (‘pDR-1500,’ Thermo Scientific,

Franklin, MA); CO, EasyLog CO-300 (Lascar Electronics Ltd., Erie, PA); O3, Personal

Ozone Monitor (‘POM,’ 2BTechnologies, Boulder, CO); and noise, Spark 703+ (Larson-

Davis Inc., Depew, NY). The reference DRIs used in this study for each hazard are

shown in Table 4.1 alongside the low-cost sensors in the sensor network.

Network-Derived Exposure Estimates

Our method for computing sensor network-derived exposure estimates is depicted in

Figure 4.1, where two pieces of information are integrated: 1) the location of the

simulated worker and 2) hazard concentration/level at the position of interest. The blue

line indicates the route traveled by a simulated employee and the blue “X” indicates the

90

location of the employee at a given time (t0, t1… tn). Hazard concentrations/levels are

represented by hazard maps, and for each time of interest, t, we estimated the hazard

concentration/level, at the location of interest, (x,y). We used an inverse distance

weighting scheme to interpolate hazard levels at unmeasured locations, which are

displayed as a hazard maps for each 5-min period. Location for the stationary routine was

taken as the coordinate where the suite of DRIs was deployed and did not change for the

duration of the sampling period. Location and time information for the mobile sampling

routine was recorded at every movement of study staff as they traveled throughout the

facility, generally following a pattern of walking 1-2 min (24-41 m), remaining stationary

for 5-15 min and walking again to the next location. Because the sensor network records

hazard measurements every 5 minutes, we constructed hazard estimates at 5-minute

intervals also (e.g. t0 = 7:00, t1 = 7:05… tn = 7:00 + 0:05∙n). All data analysis was

performed with MATLAB R2017a (Natick, MA).

Data Analysis & Comparing Network-Derived Estimates and DRI Measurements

All analyses were performed on paired 5-min data from the network-derived exposure

estimates and DRI measurements. For each simulated work shift we plotted the network-

derived exposure estimate timeseries with the DRI measurement timeseries to

qualitatively assess their overall agreement and correlation. The bias between paired

network-derived exposure estimates and DRI measurements, B, was calculated according

to:

𝐵 =𝜇

𝐶𝑇− 1 (1)

91

where μ is the network-derived exposure estimate and CT is the “true concentration” of

the hazard level measured by the reference DRI (NIOSH 2012). We expressed all

calculated biases as a percent. Adjacent to each timeseries comparing network-derived

exposure estimates and DRI measurements, we also plotted empirical cumulative density

function (CDF) curves of the bias of network-derived estimates with respect to reference

DRI measurements. Perfect agreement between network-derived estimates and reference

DRI measurements would be represented by a vertical line from zero to one at bias equal

to zero. In practice, a steeper rise in the CDF curve around percent bias equal to zero to a

value of one indicates better agreement between the network-derived exposure estimates

and reference DRI measurements. In a similar fashion to the CDFs presented next to the

timeseries of network-derived estimates and reference DRI measurements, we created

CDF curves displaying the bias between the network-derived exposure estimates and the

reference DRI measurements for August 2017, December 2017, March 2018 and those

three periods combined.

We pooled the 5-min pairs of network-derived exposure estimates and DRI

measurements by simulated employee type (stationary or mobile) and by the month

collected (August 2017, December 2017, March 2018, and August, December and March

combined) for the following bias, agreement and correlation computations. We calculated

the magnitude of the median percent bias between pairs of network-derived exposure

estimates and DRI measurements. We tabulated a measure of agreement between

network-derived estimates and reference DRI measurements by calculating the fraction of

92

network-derived exposure estimates that were within (±) 10, 25, 50 and 100% of the

reference DRI measurements. Lastly, we calculated the Pearson’s correlation coefficient

between the network-derived exposure estimates and the DRI measurements.

RESULTS

Examples of timeseries and bias CDFs for the August 2017 sampling period comparing

network-derived mobile exposure estimates to reference DRI measurements are shown in

Figure 4.2. The results of all field sampling and personal estimates are summarized in

Table 4.2 (stationary routine) and Table 4.3 (mobile routine). We collected data for three

stationary and one mobile routine on one day in August 2017, eight stationary and two

mobile routines over two days in December 2017, and eight stationary and two mobile

routines over two days in March 2018. For each routine we paired 5-min network-derived

exposure estimates with 5-min DRI measurements and tabulated the number of pairs for

August 2017, December 2017, March 2018, and those three periods combined. The

number of 5-min pairs differed between hazards and time periods due to instrument

allocation, run times and equipment failures. For example, in December 2017 for eight

stationary routines of approximately six hours each we collected a total of 553 pairs of 5-

min network-derived exposure estimates and DRI measurements for both CO and noise.

In comparison, we collected 351 pairs for PM due to number of PM DRIs available and

only 180 pairs for O3, due to DRI failures.

The median hazard levels of PM, CO, O3 and noise measured by the reference DRIs

varied between each of the sampling periods (August 2017 vs. December 2017 vs. March

93

2018), are shown in Table 4.2 (stationary routine) and Table 4.3 (mobile routine), and

generally reflected manufacturing activity in the facility. Specifically, production in the

facility during the December 2017 period was low due to the upcoming holiday

shutdown, and comparatively high for the March 2018 period. For example, the lowest

PM concentrations were observed for both stationary and mobile reference DRI

measurements in December 2017 (median PM concentrations: stationary = 0.23 mg/m3;

mobile = 0.28 mg/m3) and were highest in March 2018 (median PM concentrations:

stationary = 0.670 mg/m3; mobile = 0.544 mg/m3). Other hazards displayed similar

patterns but were not as clear as PM. For instance, O3 concentrations were low in

December 2017 (median O3 concentrations: stationary = 27 ppb; mobile = 29 ppb),

although slightly lower in August 2017, but were markedly higher in March 2018

(median O3 concentrations: stationary = 107 ppb; mobile = 124 ppb).

Comparison of stationary routine measurements and network-derived exposure estimates

are shown in Table 4.2. The number of 5-min pairs of network-derived exposure

estimates and DRI measurements, N, ranged between 84 (PM, August 2017) and 772

(CO, March 2018). The magnitude of the median biases varied by hazard. The observed

magnitude of the combined median biases for PM, CO, O3 and noise was 32, 23, 141, and

1%, respectively. For the stationary routine, the fraction of measurements within a given

percentage of reference DRIs was highest for noise, for example, with all combined

network-derived exposure estimates falling within 10% of reference DRIs. In

comparison, 0.20, 0.20 and 0.09 of combined network-derived estimates were within

10% of DRI measurements for PM, CO and O3, respectively. Correlation between

94

network-derived estimates and DRI measurements varied for each hazard as well, and for

the combined time period, the Pearson’s correlation coefficient, r, was equal to 0.48 for

PM, 0.61 for CO, 0.67 for O3, and 0.75 for noise. However, for some specific sampling

periods, the correlation was much higher than the combined period, for example PM in

August 2017 was equal to 0.81, CO in December 2017 was equal to 0.84, and O3 in

August 2017 was 0.8. The CDFs of the bias between network-derived exposure estimates

and the reference DRI measurements are displayed graphically in Figure 4.3. Compared

to the fraction of network-derived estimates within 10, 15, 50, and 100% of the reference

DRI presented in Table 4.2, both negative and positive bias are displayed in the CDF

plots in Figure 4.3.

Results for all mobile routine DRI measurements and network-derived exposure

estimates are presented in Table 4.3. The number of 5-min pairs of network-derived

exposure estimates and DRI measurements, N, ranged between 55 (all hazards, August

2017) and 156 (CO and O3, December 2017). In the mobile routine we observed the

magnitude of the median biases between network-derived estimates and DRI

measurements varied by hazard. For PM, the overall magnitude of the median bias was

36%, although we observed variation between August 2017, December 2017 and March

2018. For CO, the magnitude of the median bias increased from 15 to 20% from August

2017 to March 2018. We observed the highest magnitude of the median bias for the

combined period for O3, which decreased over time, from 369% in August 2017 to 31%

in March 2018. For noise, the magnitude of the median bias for August 2017 and

December 2018 was 1% and 4% respectively. In the mobile routine, we observed the

95

largest fraction of network-derived noise estimates were within a given percent of the

DRI measurement. For example, 0.96 of the combined noise estimates were within 10%

of the reference DRI, compared to 0.25 for CO, 0.14 for PM and 0.06 for O3. The

correlation observed between each of the network-derived hazard estimates and their

respective DRIs also varied for the mobile routine. The combined correlation was highest

for CO (r = 0.66), whereas the lowest was for noise (r = 0.39). Unfortunately, due to

equipment failure in the March 2018 sampling period, no mobile noise DRI

measurements were collected. The bias between network-derived mobile exposure

estimates and reference DRI measurements are presented graphically in Figure 4.4.

Generally, we observed similarity between the stationary and personal routines with

respect to the fraction of network-derived exposure estimates that fell within 10, 15, 50,

and 100% of their corresponding reference DRI measurements. Of all hazards, the

combined network-derived exposure estimates for noise had the largest fraction of

estimates within the smallest percent of the reference DRI for both the personal and

stationary routines, where 0.96 – 1.00 of network-derived exposure estimates were within

10% of reference DRI measurements. In contrast, the fraction of network-derived

exposure estimates for O3 within 100% of the reference DRI was equal to 0.45 for the

combined stationary routine and 0.47, for the combined mobile routine.

The correlation between network-derived exposure estimates and reference DRI

measurements varied by hazard, stationary versus mobile routine and time period. For

example, by hazard, the combined time periods of the mobile routines the correlation was

96

highest for CO (r = 0.66) and lowest for noise (r = 0.39). Variability in correlation by

routine is demonstrated with the combined stationary time periods, with the highest

correlation for noise (r = 0.75) and the lowest correlation for PM (r = 0.48). An example

of variability in correlation by sampling period, was observed for the mobile O3 routine –

in December 2017, the correlation coefficient was equal to -0.05, compared to August

2017 where it was equal to 0.63. These differences in correlation between hazards and

stationary versus mobile routine could have been affected by the by the range and

variability of the hazards during each study period and routine. For example, according to

reference DRIs, the combined median ± IQR (and range, not shown in tables) for CO was

equal to 6 ± 3 ppm (14 ppm) for the stationary routine and 5 ± 2 ppm (20 ppm) for the

mobile routine; while noise had a combined mean ± IQR equal to 81 ± 2 dBA (27 dBA)

for the stationary routine and 82 ± 3 dBA (18 dBA) for the mobile routine. The larger the

range of the hazard and the more evenly data are distributed across that range may result

in higher correlation coefficients observed for some hazards and periods of time than

others.

DISCUSSION

The success of our approach to estimating personal exposure highly depends on the

accuracy of the underlying hazard measurements of the sensor network. We have

previously reported on the accuracy of this sensor network’s measurements by

conducting experiments where each monitor was collocated with reference DRIs for one

to five minutes (Zuidema et al. in revision). In that study over a range of hazard levels,

we observed that the noise sensor in our network had the lowest median percent bias with

97

respect to the reference DRI (1%), whereas to PM had the highest median percent bias

(41%). In the same study, we observed bias as high as 524% at lower concentrations of

PM and O3.

While the sources of measurement error from low-cost sensors differ, they can often be

attributed to issues of sensitivity and specificity, in part due to sensor drift or degradation

over time or responsiveness to non-target species. The PM sensor in our network showed

evidence of decreasing sensitivity over time due to sensor loading or fouling (Thomas et

al. 2018). Furthermore, the PM sensors may produce signals that vary with different PM

composition or size distribution (Sousan et al. 2016). Our network is constructed with an

oxidative gas sensor to estimate O3 concentrations. In addition to O3 the sensor also

responds to NO2 without discrimination (Hossain et al. 2016; Spinelle et al. 2015a),

complicating the estimation of O3 in environments where NO2 is also present. Future

work may be able to incorporate the manufacturer’s proposed method to pair an oxidizing

gas sensor with and NO2-specific sensor to improve the accuracy of O3 measurements

(Hossain et al. 2016). Another source of error in this study is a ceiling observed on the

CO sensor as configured in our network at approximately 12 ppm CO (Afshar-Mohajer et

al. 2018), resulting from a the optimization of the CO sensor signal for concentrations

anticipated at the study site (Thomas et al. 2018). Because of this, the CO sensors are not

sensitive to increasing CO concentrations above 12 ppm (totaling approximately 1% of

all CO DRI measurements in this study). These errors in measurement translate to

potential errors in estimating personal exposure. In contrast, the noise sensor, which was

designed specifically for this sensor-network (Hallett et al. 2018), did not show evidence

98

of signal drift or degradation over time, and provided network-derived estimates with the

smallest bias with respect to the DRI measurements. However, despite the noise sensor

having the lowest bias, in some sampling periods, the correlation coefficient was low (r =

0.23 for the August 2017 mobile routine), demonstrating that correlation between a

sensor and reference DRI may not best measure of sensor performance.

Our approach of estimating exposure requires utilizing hazard measurements from a low-

cost sensor network with high temporal and spatial resolution. Despite these challenging

requirements, in this study we demonstrate it is feasible. The sensor network time

resolution was five minutes; accordingly, we used a five-minute averaging time for the

reference DRIs for comparison. Although this temporal resolution was high compared to

shift-long time-weighted averages (TWAs), our approach was incapable of finer time

resolution and may fail to accurately capture the peaks of brief high exposure events,

especially for hazards that decay quickly, for example, impact or impulse noise. Another

example is O3, which is highly reactive and degrades quickly after it is produced.

Limitations in temporal resolution especially affects the estimates for employees that

move through the facility at a rapid pace potentially covering large distances in the

facility in five minutes, such as materials handlers or forklift operators. We were unable

to simulate these types of rapidly moving employees in this study. To estimate hazard

levels at locations where nodes of the sensor network were not located we interpolated

hazards at unmeasured locations using an inverse-distance weighting (IDW) scheme.

While this spatial interpolation undoubtedly introduced some degree of error, the nodes

of our network were spatially dense, with the maximum distance to the nearest monitor

99

equal to 40 m (135 ft), helping to avoid errors related to spatial interpolation. Still, the

potential to mischaracterize the spatial variability of hazards, especially those that

decrease rapidly from their sources remains.

In this study for simulated mobile employees, location information was supplied by study

staff keeping a location diary during the sampling period. Although this was necessary to

demonstrate our approach for generating network-derived exposure estimates, it is not

practical for employees/employers. While previous exposure assessment studies have

used Global Positioning Systems (GPS) successfully (Adams et al. 2009; Beekhuizen et

al. 2013), unfortunately, they generally perform poorly indoors due to interference from

building roofs and a lack of “line-of-sight” to the satellites (Mainetti et al. 2014).

Therefore, for indoor/occupational settings, technologies specifically capable of indoor

localization are necessary. These indoor positioning systems include radio frequency

identification (RFID), wireless local area networks (WLAN), indoor GPS, and ultra-wide

band radio frequency (Huang et al. 2010; Khoury and Kamat 2009; Sakata et al. 2002),

and have been investigated in construction, manufacturing, warehouses, agriculture and

healthcare settings (Ahuja and Potti 2010; Bai et al. 2012; Khoury and Kamat 2009; Lim

et al. 2013; Liu et al. 2007; Sharma et al. 2012). Future work will focus on the use of

indoor positioning systems such as these to provide location information for generating

exposure estimates derived from a low-cost sensor network in an occupational setting.

Despite these challenges, this study had many novel features and strengths. This is the

first example that we are aware of that used a sensor network to estimate personal

exposure in an occupational environment. The sensor network achieved a high degree of

spatial resolution, reducing errors related to spatial interpolation. We were able to

100

estimate exposures at a relatively high temporal resolution, also a benefit over shift-long

TWAs. We maintained a high degree of accuracy for the location information on

simulated mobile employees with respect to both time and space with position diaries.

Consequently, the location information we used to estimate personal exposures did not

have errors that would have been inherent to those provided by in an indoor positioning

system. Our multi-hazard sensor network was deployed at the study site continuously for

nearly eight months. While we only had access to the facility for five days over that

period to conduct personal sampling, we demonstrated the ability of our technique to

potentially provide personal exposure estimates for any employee whose position can be

tracked over that time. This kind of information on individual workers would be a vast

improvement over traditional personal sampling rates (Roick et al. 1991; Tornero‐Velez

et al. 1997) with beneficial implications for both occupational exposure assessment for

OEL compliance and epidemiological study.

101

TABLES & FIGURES

Table 4.1. Low-cost sensors and Reference DRIs used to measure occupational hazards.

Hazard Network Sensor Reference DRI

PM GP2Y1010AU0F (SHARP Electronics, Osaka, Japan) pDR-1500 (Thermo Scientific, Franklin, MA)

CO CO-B4 (Alphasense Ltd., Essex, UK) EasyLog CO-300 (Lascar Electronics Ltd., Erie, PA)

O3 OX-B431 (Alphasense Ltd., Essex, UK) POM (2BTechnologies, Boulder, CO)

Noise Custom (Hallett et al. 2018) Spark 703+ (Larson-Davis Inc., Depew, NY)

102

Table 4.2. Comparison of reference DRI measurements and network-derived exposure estimates (pairs of 5-minute averages)

for the stationary routine.

Fraction within Percent of DRIA

Hazard Time Period

# Simulated

Work Shifts

# 5-min

Pairs, N

DRI Median

(IQR)

Median

|%Bias| (IQR) 10 25 50 100

Correlation

with DRI

PM DRI units: mg/m3

Aug-2017 3 84 0.28 (0.10) 11 (16) 0.45 0.86 1.00 1.00 0.81

Dec-2017 8 351 0.23 (0.12) 106 (294) 0.10 0.20 0.34 0.50 0.31

Mar-2018 8 380 0.67 (0.42) 22 (43) 0.23 0.54 0.74 0.81 0.46

Combined 19 815 0.31 (0.40) 32 (143) 0.20 0.43 0.60 0.69 0.48

CO DRI units: ppm

Aug-2017 3 207 8 (4) 28 (13) 0.12 0.39 0.94 0.98 66

Dec-2017 8 553 6 (1) 14 (11) 0.34 0.95 1.00 1.00 0.84

Mar-2018 8 772 7 (4) 36 (34) 0.12 0.29 0.62 0.90 0.56

Combined 19 1532 6 (3) 23 (27) 0.20 0.54 0.80 0.94 0.61

O3 DRI units: ppb

Aug-2017 3 204 27 (23) 325 (260) 0.00 0.00 0.00 0.00 0.8

Dec-2017 8 180 29 (8) 254 (144) 0.00 0.00 0.00 0.03 0.62

Mar-2018 8 664 107 (91) 42 (126) 0.15 0.34 0.55 0.69 0.56

Combined 19 1048 56 (95) 141 (220) 0.09 0.22 0.35 0.45 0.67

Noise DRI units: dBA

Aug-2017 3 207 82 (2) 2 (1) 1.00 1.00 1.00 1.00 0.65

Dec-2017 8 553 80 (3) 1 (2) 1.00 1.00 1.00 1.00 0.77

Mar-2018 8 634 81 (2) 1 (1) 0.99 1.00 1.00 1.00 0.65

Combined 19 1394 81 (2) 1 (1) 1.00 1.00 1.00 1.00 0.75

A Fraction of network-derived estimates that are within (±) 10, 25, 50 and 100% of the direct-reading instrument (DRI) measurements

for each hazard.

103

Table 4.3. Comparison of reference DRI measurements and network-derived exposure estimates (pairs of 5-minute averages)

for the mobile routine. Equipment failure resulted in no personal noise measurements in March 2018.

Fraction within Percent of DRIA

Hazard Time Period

# Simulated

Work Shifts

# 5-min

Pairs, N

DRI Median

(IQR)

Median

|%Bias| (IQR) 10 25 50 100

Correlation

with DRI

PM DRI units: mg/m3

Aug-2017 1 55 0.45 (0.33) 26 (22) 0.20 0.45 0.89 1.00 0.77

Dec-2017 2 153 0.28 (0.17) 52 (73) 0.10 0.28 0.48 0.77 0.07

Mar-2018 2 154 0.54 (0.30) 34 (69) 0.16 0.39 0.6 0.82 0.11

Combined 5 362 0.38 (0.30) 36 (65) 0.14 0.35 0.60 0.83 0.43

CO DRI units: ppm

Aug-2017 1 55 7 (5) 15 (23) 0.25 0.64 0.91 0.98 0.86

Dec-2017 2 156 5 (2) 18 (21) 0.25 0.63 0.90 1.00 0.59

Mar-2018 2 153 4 (2) 20 (30) 0.25 0.58 0.80 0.92 0.41

Combined 5 364 5 (2) 18 (25) 0.25 0.61 0.86 0.96 0.66

O3 DRI units ppb

Aug-2017 1 55 23 (25) 369 (758) 0.00 0.00 0.00 0.00 0.63

Dec-2017 2 91 29 (10) 233 (168) 0.00 0.00 0.00 0.03 -0.05

Mar-2018 2 155 124 (97) 31 (34) 0.12 0.39 0.74 0.90 0.54

Combined 5 301 53 (99) 119 (228) 0.06 0.20 0.38 0.47 0.52

Noise DRI units: dBA

Aug-2017 1 55 83 (2) 1 (2) 0.96 1.00 1.00 1.00 0.23

Dec-2017 2 156 82 (3) 4 (4) 0.96 1.00 1.00 1.00 0.43

Mar-2018 0 0 -- -- -- -- -- -- --

Combined 3 211 82 (3) 3 (4) 0.96 1.00 1.00 1.00 0.39

A Fraction of network-derived estimates that are within (±) 10, 25, 50 and 100% of the direct-reading instrument (DRI) measurements

for each hazard.

104

Figure 4.1. Schematic of technique to estimate personal exposure from sensor network. Personal estimates are derived by

taking the hazard concentration/level at location (x,y) for time t.

105

Figure 4.2. Examples of timeseries comparing network-derived exposure estimates (dashed line) with reference DRI

measurements (solid line) for simulated mobile employees for a) PM, b) CO, c) O3, and d) noise. Cumulative Density Function

(CDF) of the bias of network-derived exposure estimates with respect to the reference DRI measurement shown adjacent to

each timeseries.

Figure 4.3. CDF plots of for the stationary routine for a) PM, b) CO, c) O3, and d)

noise displaying the fraction of network-derived exposure estimates and bias with

respect to reference DRI measurements. The blue curve is for August 2017, the

orange curve is for December 2017, the yellow curve is for March 2018, and the bold

black curve is for all sampling periods combined.

107

Figure 4.4. CDF plots of for the mobile routine for a) PM, b) CO, c) O3, and d) noise

displaying the fraction of network-derived exposure estimates and bias with respect

to reference DRI measurements. The blue curve is for August 2017, the orange

curve is for December 2017, the yellow curve is for March 2018, and the bold black

curve is for all sampling periods combined.

108

REFERENCES

Adams C, Riggs P, Volckens J. 2009. Development of a method for personal,

spatiotemporal exposure assessment. Journal of Environmental Monitoring 11:1331-

1339.



ozone, nitrogen dioxide, and carbon monoxide. Journal of Occupational and

Environmental Hygiene 15:87-98.

Ahuja S, Potti P. 2010. An introduction to rfid technology. Communications and Network

2:183.

Anderson JO, Thundiyil JG, Stolbach A. 2012. Clearing the air: A review of the effects

of particulate matter air pollution on human health. Journal of Medical Toxicology

8:166-175.

Bai YB, Wu S, Wu HR, Zhang K. Overview of rfid-based indoor positioning technology.

In: Proceedings of the GSR, 2012 2012, Citeseer.

Beekhuizen J, Kromhout H, Huss A, Vermeulen R. 2013. Performance of gps-devices for

environmental exposure assessment. Journal of Exposure Science and Environmental

Epidemiology 23:498-505.




109

Bornholdt J, Dybdahl M, Vogel U, Hansen M, Loft S, Wallin H. 2002. Inhalation of

ozone induces DNA strand breaks and inflammation in mice. Mutation

Research/Genetic Toxicology and Environmental Mutagenesis 520:63-72.

Dockery DW. 1993. Epidemiologic study design for investigating respiratory health

effects of complex air pollution mixtures. Environmental health perspectives 101

Suppl 4:187-191.




125.

Evans DE, Heitbrink WA, Slavin TJ, Peters TM. 2008. Ultrafine and respirable particles

in an automotive grey iron foundry. The Annals of Occupational Hygiene 52:9-21.




Hallett L, Tatum M, Thomas G, Sousan S, Koehler K, Peters T. 2018. An inexpensive

sensor for noise. J Occup Environ Hyg 15:448-454.



computing, v. 16 n.



110


19.

Heitbrink WA, Evans DE, Peters TM, Slavin TJ. 2007. Characterization and mapping of

very fine particles in an engine machining and assembly facility. Journal of




Huang F-C, Shih T-S, Lee J-F, Chao H-P, Wang P-Y. 2010. Time location analysis for

exposure assessment studies of indoor workers based on active rfid technology.

Journal of Environmental Monitoring 12:514-523.



1:1.






5292.

Kampa M, Castanas E. 2008. Human health effects of air pollution. Environmental


111

Khoury HM, Kamat VR. 2009. Evaluation of position tracking technologies for user

localization in indoor construction environments. Automation in Construction

18:444-457.



840.



2610.







Lim MK, Bahr W, Leung SCH. 2013. Rfid in the warehouse: A literature analysis (1995–

2010) of its applications, benefits, challenges and future trends. International Journal

of Production Economics 145:409-430.

Lippmann M. 1989. Health effects of ozone a critical review. JAPCA 39:672-695.

Liu H, Darabi H, Banerjee P, Liu J. 2007. Survey of wireless indoor positioning

techniques and systems. IEEE Transactions on Systems, Man, and Cybernetics, Part

C (Applications and Reviews) 37:1067-1080.

112

Liu S, Hammond SK. 2010. Mapping particulate matter at the body weld department in

an automobile assembly plant. Journal of occupational and environmental hygiene

7:593-604.

Mainetti L, Patrono L, Sergi I. A survey on indoor positioning systems. In: Proceedings

of the Software, Telecommunications and Computer Networks (SoftCOM), 2014

22nd International Conference on, 2014 2014, IEEE, 111-120.















OSHA 1974. OSHA noise regulations (standards-29 cfr): Occupational noise exposure.-

1910.95.

113

OSHA 1993. OSHA safety and health standards. 29 cfr 1910.1000. Table z-1. 1:444.

Ott DK, Kumar N, Peters TM. 2008. Passive sampling to capture spatial variability in pm

10–2.5. Atmospheric Environment 42:746-756.

Park JY, Ramachandran G, Raynor PC, Olson Jr GM. 2010. Determination of particle

concentration rankings by spatial mapping of particle surface area, number, and mass

concentrations in a restaurant and a die casting plant. Journal of occupational and

environmental hygiene 7:466-476.

Passchier-Vermeer W, Passchier WF. 2000. Noise exposure and public health.

Environmental health perspectives 108 Suppl 1:123-131.




Peters TM, Anthony TR, Taylor C, Altmaier R, Anderson K, O'Shaughnessy PT. 2012.

Distribution of particle and gas concentrations in swine gestation confined animal

feeding operations. The Annals of Occupational Hygiene 56:1080-1090.




Pope CA, Dockery DW, Schwartz J. 1995. Review of epidemiological evidence of health

effects of particulate air pollution. Inhalation toxicology 7:1-18.

Pope III CA, Dockery DW. 2006. Health effects of fine particulate air pollution: Lines

that connect. Journal of the Air & Waste Management Association 56:709-742.

114

Ramachandran G. 2008. Toward better exposure assessment strategies--the new niosh

initiative. The Annals of Occupational Hygiene 52:297-301.



Rappaport SM, Kupper LL. 2008. Quantitative exposure assessment:S. Rappaport.

Raub JA, Mathieu-Nolf M, Hampson NB, Thom SR. 2000. Carbon monoxide

poisoning—a public health perspective. Toxicology 145:1-14.

Rezagholi M, Mathiassen SE. 2010. Cost-efficient design of occupational exposure

assessment strategies--a review. The Annals of Occupational Hygiene 54:858-868.

Roick JC, Norwood SK, Hawkins NC. 1991. A strategy for occupational exposure

assessment:AIHA.

Sakata M, Yasumuro Y, Imura M, Manabe Y, Chihara K. A location awareness system

using wide-angle camera and active ir-tag. In: Proceedings of the MVA, 2002 2002,

522-525.

Sferlazza SJ, Beckett WS. 1991. The respiratory health of welders1-3. Am Rev Respir

Dis 143:1134-1148.

Sharma D, Thomas GW, Foster ED, Iacovelli J, Lea KM, Streit JA, et al. 2012. The

precision of human-generated hand-hygiene observations: A comparison of human

observation with an automated monitoring system. Infection Control & Hospital


115




Sousan S, Koehler K, Thomas G, Park JH, Hillman M, Halterman A, et al. 2016. Inter-

comparison of low-cost sensors for measuring the mass concentration of occupational

aerosols. Aerosol Science and Technology 50:462-473.

Spinelle L, Gerboles M, Aleixandre M. 2015. Performance evaluation of amperometric





Tornero‐Velez R, Symanski E, Kromhout H, Yu RC, Rappaport SM. 1997. Compliance

versus risk in assessing occupational exposures. Risk Analysis 17:279-292.

Vosburgh DJH, Boysen DA, Oleson JJ, Peters TM. 2011. Airborne nanoparticle

concentrations in the manufacturing of polytetrafluoroethylene (ptfe) apparel. Journal

of occupational and environmental hygiene 8:139-146.

Weschler CJ. 2006. Ozone's impact on public health: Contributions from indoor

exposures to ozone and products of ozone-initiated chemistry. Environmental health

perspectives:1489-1496.

Zuidema C, Sousan S, Stebounova LV, Gray A, Liu X, Tatum M, et al. in revision.

Mapping occupational hazards with a multi-sensor network in a heavy-vehicle

manufacturing facility. Annals of Work Exposure and Health.

116

CHAPTER FIVE

Conclusion

117

SUMMARY FINDINGS

Aim 1. Evaluate a low-cost sensor solution for quantifying NO2 and O3

concentrations in mixture.

In laboratory calibration procedures, we observed the individual response of the

Alphasense NO2-B43F sensors to NO2 and OX-B431 sensors to NO2 and O3 were highly

linear over the concentrations studied (R2 ≥ 0.99), but the slopes of the response for each

sensor were unique, suggesting sensor-specific calibration was prudent. As expected, the

NO2-B43F sensor did not respond to O3 gas. In mixtures of NO2 and O3, the absolute

mean percent bias was much larger for O3 (between -187 and -24%) compared to NO2

(between -8 and 29%). We observed instability of the senor baseline over 4 days of

experiments equivalent to 34 ppb O3, prompting an alternate method of baseline-

correcting sensor signal to calculate concentrations. The baseline-correction method

resulted in mean percent bias between -44 and 17% for NO2 and between -107 and 5%

for O3. Both analysis methods progressively underestimated O3 concentrations as the

ratio of NO2 concentration to O3 concentration increased. Our results suggested that these

paired electrochemical sensors are selective for O3 in mixture with NO2, but that O3

concentration estimates are subject to degrees of error that make their use challenging.

118

Aim 2. Establish sensor networks as useful tools for measuring occupational

hazards with a high degree of space-time resolution

At the five-month point in the eight-month long deployment of our multi-hazard sensor

network at the study site, we reported on the space-time measurements from the network,

precision of network measurements, and accuracy of network measurements with respect

to field reference instruments. We observed clear diurnal and weekly temporal patterns

for all hazards and daily, hazard-specific spatial patterns attributable to general

manufacturing processes in the facility. Network sensors exhibited varying degrees of

precision with 95% of measurements among three collocated nodes within 0.23 mg/m3

for PM, 0.4 ppm for CO, 7 ppb for oxidizing gases, and 1 dBA for noise of each other.

The median percent bias with reference to DRIs varied by hazard and was equal to 41%

for PM, 7% for CO, 36% for ozone (measured by the oxidizing gas sensor) and 1% for

noise. Our network allowed us to measure multiple hazards simultaneously across the

study site and create hazard maps for each of the hazards for any time period of interest

during a continuous eight-month-long deployment. These features were a substantial

improvement over prior methods in hazard mapping, which typically involved traversing

a facility with DRIs over a limited period of time. Additionally, with a sensor network,

the measurement errors associated with interpolating hazard measurements over time and

space are greatly reduced compared to what is practical mapping hazards with higher-

quality DRIs. Another novel feature of our sensor network was that the 40 nodes were

deployed at the study site in a spatially optimized pattern that captured a maximum

amount of spatial variability compared to grid or random monitor placement (Berman et

119

al. 2018). In this aim we demonstrated that sensor networks can be successfully used to

provide long-term hazard data in an industrial manufacturing setting with a high level of

temporal and spatial resolution but are still limited by the sensitivity and specificity of

sensors.

Aim 3. Develop a method to estimate personal exposure to occupational hazards and

compare traditional personal measurements to network-derived estimates.

At three different points during the eight months the multi-hazard sensor network was

deployed, we conducted fieldwork to simulate both stationary and mobile workers at the

study site and estimated personal exposure with the sensor network. To estimate personal

exposures, we integrated the hazard data from the sensor network with location

information on two types of simulated workers. The first type of simulated worker was

one that remained in a relatively small geographic space to perform their work duties,

such as a welder or machine operator. The second type of simulated worker was one that

was highly mobile and traveled throughout the facility at a slow pace (e.g. walking), such

as supervisors, mechanics, and maintenance workers. Location information was recorded

by study staff in a diary according to an established coordinate system at the study site,

marked by regularly spaced structural I-beams. To assess the accuracy of the network-

derived exposure estimates, we took typical industrial hygiene personal measurements

using reference DRIs and compared these exposure measurements to corresponding

exposure estimates. Network-derived exposure estimates compared favorably to personal

measurements taken with a suite of reference DRIs but varied by hazard and type of

120

simulated employee. The median percent bias between network-derived exposure

estimates and DRI measurements for simulated stationary employees was 32% for PM,

23% for CO, 141% for O3, and 2% for noise; and for simulated mobile employees was

36% for PM, 18% for CO, 119% for O3, and 3% for noise. Correlation between network-

derived exposure estimates and DRI measurements varied greatly among between the

hazards and type of simulated employee.

FUTURE RESEARCH, PUBLIC HEALTH IMPLICATIONS, AND

CONCLUDING REMARKS

Future Research

This dissertation highlights two main areas for future research. First, in our study of low-

cost electrochemical sensors, we contributed to evidence that the strategy of filtering out

cross sensitive gases is effective – the NO2-B4 sensors did not exhibit a response to O3.

However, in the case of NO2 and O3, the industry solution we evaluated still struggled to

overcome the challenge of detecting O3 concentrations in the part-per-billion range when

NO2 concentrations in the parts-per-million range. Using a similar strategy of deploying

collocated sensors for other target gases with known interferents may be successful

though, depending on the concentrations of interest of the gases and the required

accuracy of the application. Pairing sensors has the potential to address a major limitation

of low-cost sensors, namely their response to non-target species. Opportunities to reduce

extraneous sensor response will reduce measurement errors associated with low-cost

121

sensors and will improve their accuracy and utility for environmental health research and

air pollution monitoring.

Second, while we were limited to study staff keeping a location diary, we identified the

opportunity for automated indoor positioning systems to provide location information for

estimating personal exposure from a sensor network. Unfortunately, well known global

positioning systems (GPS) generally do not perform well indoors because of interference

from building roofs and a lack of “line-of-sight” to the satellites (Mainetti et al. 2014).

Consequently, for indoor/occupational settings, technologies specifically capable of

indoor localization must be used. A variety of indoor positioning systems are

commercially available including radio frequency identification, wireless local area

networks, indoor GPS, and ultra-wide band radio frequency (Huang et al. 2010; Khoury

and Kamat 2009; Sakata et al. 2002), and have been investigated in construction,

manufacturing, warehouses, agriculture and healthcare settings (Ahuja and Potti 2010;

Bai et al. 2012; Khoury and Kamat 2009; Lim et al. 2013; Liu et al. 2007; Sharma et al.

2012). Future work utilizing automated indoor positioning systems such as these to

provide location information will greatly improve the estimation of personal exposure to

occupational hazards with a sensor network that we have demonstrated in this

dissertation. In this scenario, generalizability is also improved as the number of

positioning devices and willing participants are the only limiting factors, allowing

exposures to be estimated for all workers on all work days.

122

Public Health Implications

The current system of occupational exposure assessment, relying on infrequent personal

measurements may leave workplace exposures inadequately characterized and risks

higher than measurements indicate (Rappaport 1984). While some researchers have

documented the challenge of characterizing variability between- and within-workers,

their recommendations still rely on conducting personal measurement (Kromhout et al.

1993; Rappaport et al. 1993). In this dissertation we offer a different strategy to provide

more exposure data, without necessarily increasing the amount of personal sampling

required. We show that hazard measurements from sensor networks offer rich datasets to

comprehensively assess occupational hazards and estimate personal exposure. Despite the

error observed estimating personal exposure to some of the hazards under study,

estimating personal exposure holds promise as an additional tool to be used with

traditional personal sampling. If errors associated with low-cost sensors can be reduced

further, this strategy may even provide an alternative to personal sampling. Due to the

ability to frequently, easily, and simultaneously collect exposure information on many

workers, estimating personal exposure has the potential to offer a wealth of information

to examine hazards and reduce occupational risk.

Beyond the opportunities for compliance and occupational exposure monitoring, sensor

networks, such as ours could also be used to provide improved exposure data for

occupational epidemiology studies. Exposure assessment is often viewed as a weak link

in epidemiological studies, commonly relying on crude exposure classifications, indirect

123

exposure data, historically reconstructed exposure assessments, or exposure assessments

with substantial measurement error. As we have demonstrated here, personal exposure to

a variety of occupational hazards can be estimated simultaneously for many workers with

a sensor network, a potentially powerful tool for occupational epidemiology. Future

research that takes advantage of the ability to better characterize exposure variability may

lead to improved understanding of risk factors associated with work-related health

effects.

This dissertation also offers lessons for environmental health scientists and researchers

who do not work in the occupational setting but are interested in using sensors to provide

exposure measurements or augment regulatory monitoring networks. For example, we

show the range of hazard levels that can reasonably be measured with different sensors,

the precision associated with collocated sensors, the accuracy that can be achieved with

respect to higher-quality direct-reading instruments, and the calibration requirements of

different sensors, which are useful for others in their future research. Certainly, however,

researchers in the outdoor/general environment must contend with challenges that are

different than the ones we faced, such as low pollutant concentrations and the limits of

detection and greater ranges of temperature and relative humidity in the outdoor

environment that affect sensor signal.

Concluding Remarks

124

In recent years low-cost sensors have attracted the attention of environmental health

researchers (Kumar et al. 2015; Lewis et al. 2016; Masson et al. 2015; Piedrahita et al.

2014; Snyder et al. 2013), and we believe for good reason. They offer an opportunity to

characterize environmental exposures with an unprecedented degree of spatial and

temporal resolution when integrated into sensor networks. While much of the work has

up to this point focused on sensor networks as tools to monitor ambient air pollution

(English et al. 2017; Gao et al. 2015; Hasenfratz et al. 2015; Heimann et al. 2015; Ikram

et al. 2012; Jiang et al. 2016; Jiao et al. 2016; Kumar et al. 2011; Mead et al. 2013;

Moltchanov et al. 2015), we have demonstrated the utility of deploying sensor networks

in the workplace, and the unique opportunity to use them to estimate occupational

exposure.

125

REFERENCES

Ahuja S, Potti P. 2010. An introduction to rfid technology. Communications and Network

2:183.

Bai YB, Wu S, Wu HR, Zhang K. Overview of rfid-based indoor positioning technology.

In: Proceedings of the GSR, 2012 2012, Citeseer.







125.






computing, v. 16 n.




19.

126

Huang F-C, Shih T-S, Lee J-F, Chao H-P, Wang P-Y. 2010. Time location analysis for

exposure assessment studies of indoor workers based on active rfid technology.

Journal of Environmental Monitoring 12:514-523.



1:1.






5292.

Khoury HM, Kamat VR. 2009. Evaluation of position tracking technologies for user

localization in indoor construction environments. Automation in Construction

18:444-457.

Kromhout H, Symanski E, Rappaport SM. 1993. A comprehensive evaluation of within-

and between-worker components of occupational exposure to chemical agents. The

Annals of Occupational Hygiene 37:253-270.



2610.

127







Lim MK, Bahr W, Leung SCH. 2013. Rfid in the warehouse: A literature analysis (1995–

2010) of its applications, benefits, challenges and future trends. International Journal

of Production Economics 145:409-430.

Liu H, Darabi H, Banerjee P, Liu J. 2007. Survey of wireless indoor positioning

techniques and systems. IEEE Transactions on Systems, Man, and Cybernetics, Part

C (Applications and Reviews) 37:1067-1080.

Mainetti L, Patrono L, Sergi I. A survey on indoor positioning systems. In: Proceedings

of the Software, Telecommunications and Computer Networks (SoftCOM), 2014

22nd International Conference on, 2014 2014, IEEE, 111-120.






128









Rappaport SM, Kromhouta H, Symanski E. 1993. Variation of exposure between workers

in homogeneous exposure groups. The American Industrial Hygiene Association

Journal 54:654-662.

Sakata M, Yasumuro Y, Imura M, Manabe Y, Chihara K. A location awareness system

using wide-angle camera and active ir-tag. In: Proceedings of the MVA, 2002 2002,

522-525.

Sharma D, Thomas GW, Foster ED, Iacovelli J, Lea KM, Streit JA, et al. 2012. The

precision of human-generated hand-hygiene observations: A comparison of human

observation with an automated monitoring system. Infection Control & Hospital





129

APPENDICES

Appendix 2.1. Experimental Data for Method 1. Reference Instruments Sensor Pair 1 Sensor Pair 2 Sensor Pair 3 3 Sensor Pair Summary

[NO2]

(ppm)

[O3]

(ppb)

T

(°C) RH

(%)

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

%Var

[O3]

%Var

0.04 0 28 43 0.03 -21 -- -- 0.03 -21 -- -- 0.05 -29 -- -- 0.04 -24 -- -- -- --

0.13 0 28 38 0.12 -28 -3 -- 0.12 -26 -5 -- 0.13 -30 2 -- 0.12 -28 -2 -- 4 --

0.13 62 28 35 0.12 34 -4 -45 0.12 37 -5 -40 0.13 29 3 -54 0.13 34 -2 -46 5 13

0.11 124 28 34 0.10 95 -10 -23 0.09 101 -12 -19 0.11 87 0 -30 0.10 94 -8 -24 7 7

0.13 34 28 35 0.12 9 -6 -74 0.12 5 -4 -86 0.13 1 3 -97 0.13 5 -2 -86 5 83

0.13 96 28 35 0.13 58 -1 -40 0.13 64 -3 -33 0.14 55 4 -43 0.13 59 0 -38 4 8

MAPE 5 45 6 44 3 56 3 49

0.02 0 24 27 0.06 -21 -- -- 0.06 -21 -- -- 0.07 -31 -- -- 0.07 -24 -- -- -- --

0.25 0 25 37 0.27 -29 7 -- 0.28 -39 10 -- 0.27 -35 9 -- 0.27 -34 8 -- 2 --

0.26 66 26 40 0.26 42 3 -37 0.28 33 8 -51 0.28 30 8 -55 0.27 35 6 -47 3 18

0.27 130 27 39 0.28 92 2 -29 0.29 84 7 -35 0.28 83 5 -36 0.28 86 5 -34 2 6

0.28 33 27 39 0.26 1 -7 -98 0.26 -2 -6 -106 0.26 -3 -6 -110 0.26 -2 -7 -105 1 121

0.26 98 28 39 0.26 63 2 -36 0.27 56 5 -43 0.27 53 5 -46 0.27 57 4 -41 2 9

MAPE 4 50 7 59 6 62 6 57

0.04 0 28 44 0.04 -36 -- -- 0.05 -24 -- -- 0.06 -33 -- -- 0.05 -31 -- -- -- --

0.53 1 28 44 0.54 -51 2 -- 0.58 -53 9 -- 0.53 -27 1 -- 0.55 -44 4 -- 4 --

0.43 65 28 44 0.54 13 26 -80 0.58 10 35 -84 0.54 34 25 -48 0.55 19 29 -71 5 69

0.54 125 28 44 0.58 56 8 -55 0.62 63 15 -50 0.57 82 6 -34 0.59 67 9 -46 4 20

0.49 33 28 44 0.53 -35 8 -208 0.57 -29 15 -189 0.52 -4 6 -113 0.54 -23 10 -170 5 72

0.50 94 28 45 0.54 27 8 -71 0.57 33 16 -65 0.52 54 6 -42 0.54 38 10 -60 5 38

MAPE 11 104 18 97 9 59 12 87

0.03 -2 23 26 0.06 -24 -- -- 0.06 -21 -- -- 0.07 -29 -- -- 0.06 -25 -- -- -- --

1.06 2 25 36 1.14 -55 8 -- 1.23 -100 16 -- 1.11 -22 5 -- 1.16 -59 10 -- 5 --

0.96 65 26 39 1.03 24 7 -63 1.09 -11 14 -117 1.00 54 4 -17 1.04 22 8 -66 4 146

1.11 131 26 41 1.16 81 5 -38 1.24 37 12 -72 1.13 114 2 -13 1.18 77 6 -41 5 50

0.96 31 27 43 1.05 -23 9 -174 1.14 -64 18 -305 1.04 6 8 -82 1.08 -27 12 -187 5 129

1.01 97 28 43 1.10 26 9 -73 1.20 -6 19 -106 1.09 61 8 -37 1.13 27 12 -72 5 123

MAPE 8 87 16 150 5 37 10 91

Total MAPE 7 72 12 88 6 53 8 71

131

Appendix 2.2. Experimental Data for Method 2.138 Reference Instruments Sensor Pair 1 Sensor Pair 2 Sensor Pair 3 3 Sensor Pair Summary

[NO2]

(ppm)

[O3]

(ppb)

T

(°C) RH

(%)

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

(ppm)

[O3]

(ppb)

[NO2]

%Bias

[O3]

%Bias

[NO2]

%Var

[O3]

%Var

0.04 0 28 43 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- -- --

0.13 0 28 38 0.09 -7 -30 -- 0.08 -5 -33 -- 0.08 -1 -35 -- 0.08 -4 -33 -- 4 --

0.13 62 28 35 0.09 55 -30 -12 0.09 59 -32 -6 0.08 58 -34 -7 0.09 57 -32 -8 3 4

0.11 124 28 34 0.06 116 -42 -7 0.06 122 -44 -2 0.06 116 -45 -7 0.06 118 -44 -5 2 3

0.13 34 28 35 0.09 29 -32 -12 0.09 26 -31 -23 0.09 30 -34 -11 0.09 28 -32 -15 2 8

0.13 96 28 35 0.10 78 -26 -18 0.09 86 -29 -10 0.09 84 -32 -12 0.09 83 -29 -14 4 5

MAPE 32 12 34 10 36 9 34 11

0.02 0 24 27 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- -- --

0.25 0 25 37 0.21 -9 -17 -- 0.24 -24 -6 -- 0.20 -4 -21 -- 0.23 -13 -10 -- 4 --

0.26 66 26 40 0.20 63 -20 -5 0.23 47 -8 -29 0.20 61 -21 -8 0.23 56 -12 -15 3 14

0.27 130 27 39 0.22 113 -20 -13 0.25 98 -9 -24 0.21 114 -23 -13 0.24 108 -12 -17 4 8

0.28 33 27 39 0.20 21 -29 -35 0.22 12 -21 -63 0.19 28 -33 -16 0.21 20 -23 -39 4 40

0.26 98 28 39 0.20 84 -22 -14 0.23 70 -11 -28 0.19 84 -24 -14 0.22 79 -14 -19 3 9

MAPE 22 17 11 36 17 12 14 23

0.04 0 28 44 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- -- --

0.53 1 28 44 0.50 -15 -6 -- 0.53 -28 0 -- 0.48 6 -10 -- 0.50 -12 -5 -- 5 --

0.43 65 28 44 0.50 48 16 -25 0.54 35 24 -46 0.48 67 12 4 0.51 50 17 -23 5 33

0.54 125 28 44 0.53 92 0 -26 0.57 87 6 -30 0.51 116 -4 -7 0.54 98 0 -21 5 16

0.49 33 28 44 0.49 0 0 -98 0.52 -5 6 -114 0.47 29 -5 -11 0.49 8 0 -75 6 220

0.50 94 28 45 0.49 63 0 -33 0.52 57 -6 -39 0.47 88 -5 -7 0.50 69 0 -27 6 23

MAPE 5 46 8 58 7 7 5 36

0.03 -2 23 26 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- 0.00 0 -- -- -- --

1.06 2 25 36 1.09 -31 3 -- 1.17 -78 11 -- 1.05 7 -1 -- 1.10 -34 4 -- 6 --

0.96 65 26 39 0.98 48 2 -26 1.03 10 7 -85 0.93 84 -3 28 0.98 47 2 -27 5 78

1.11 131 26 41 1.10 106 0 -19 1.18 58 7 -56 1.06 143 -4 9 1.11 102 1 -22 6 42

0.96 31 27 43 1.00 1 3 -97 1.08 -43 12 -237 0.97 35 1 12 1.02 -2 6 -107 6 1753

1.01 97 28 43 1.05 50 4 -48 1.14 16 13 -83 1.02 90 1 -7 1.07 52 6 -46 6 72

MAPE 3 48 10 115 2 14 4 51

Total MAPE 13 31 16 55 17 11 14 30

132

Appendix 3.1. Pearson’s correlation between hazards and temperature.

CURRICULUM VITAE

Zuidema, Christopher

134

Christopher Matthew Zuidema

29 N Curley St, Baltimore MD 21224

Born May 29, 1986 in Ridgewood, NJ

802-829-0086 ● [email protected]

EDUCATION

PhD, Environmental Health and Engineering Expected July 2018

Concentration: Exposure Sciences and Environmental Epidemiology

Johns Hopkins University (JHU), Bloomberg School of Public Health, Baltimore, MD

• Graduate Certificate: The Food System, Environment and Public Health 2017

• Graduate Certificate: Risk Sciences and Public Policy 2016

SM, Exposure, Epidemiology and Risk, Concentration: Industrial Hygiene May 2010

Harvard University, School of Public Health, Boston, MA

BS, Science of Earth Systems May 2008

Cornell University, College of Agriculture and Life Sciences, Ithaca, NY

WORK EXPERIENCE

Student, Estimating Personal Exposure to Occupational Hazards with a Low-Cost Sensor Network

JHU Bloomberg School of Public Health, Baltimore, MD July 2016-Present

• Led research team of professors, postdocs, and graduate students

• Coordinated and conducted field and laboratory data collection, analysis

• Applied for and received additional funding from JHU Education and Research Center

Teaching Assistant JHU Bloomberg School of Public Health, Baltimore, MD August 2016-October 2017

• Prepared lectures, delivered course content, provided feedback to graduate students

o Department of Health Policy and Management

▪ Introduction to the Risk Sciences and Public Policy

▪ Risk Policy, Management and Communication

▪ Methods in Quantitative Risk Assessment

o Department of Environmental Health and Engineering

▪ Methods in the Exposure Sciences

▪ Methods in the Exposure Sciences Lab

▪ Airborne Particles

Research Assistant, Bountiful Baltimore and Maryland Food Recovery Projects

JHU Center for a Livable Future, Baltimore, MD July 2015-July 2016

• Conducted interviews with food recovery organizations about current practices and capacity

• Developed a risk assessment framework for the Bountiful Baltimore urban foraging project

Public Health Industrial Hygienist January 2011-August 2014

Vermont Department of Health (VDH), Burlington, VT

• Provided leadership and technical expertise on physical, chemical, and biological exposures

mailto:[email protected]

135

• Applied for, received, and coordinated EPA State Indoor Radon Grant

• Supervised and provided technical expertise to the VDH residential radon surveillance initiative

• Communicated technical information about radon testing and mitigation in schools and homes in

public meetings, print and television news media, advertising campaigns, and person

• Partnered with neighboring states to improve technical expertise of the VDH Radon Program

• Led public-private partnership to link VT schools with certified radon professionals for mitigation

• Organized a team of VDH experts and proposed a VT Health Guidance Value for radon in water

• Managed VDH’s EPA Tools for Schools program and responded to school concerns

• Developed and conducted environmental health walkthrough assessments in school buildings

• Performed short-term radon screenings and advised on mitigation in public school buildings

• Represented environmental health interests on the VDH Coordinated School Health Team

• Led multi-agency grant application to the EPA on biomass boilers and asthma

• Sampled indoor and outdoor air for formaldehyde from agricultural practices in coordination with

VT Agency of Agriculture and CDC ATSDR in an exposure investigation

• Coordinated VT and EPA sampling teams responding to illegal pesticide application

• Participated in VDH’s Healthy Homes Program strategic planning for CDC Healthy Homes Grant

• Evaluated Air Force environmental impact statement for F-35 basing at VT Air National Guard

• Volunteered for the VT Radiological Dose Assessment Team and trained for emergencies

Research Assistant Harvard Prevention Research Center, Boston, MA May 2009-May 2010

• Adapted EPA sampling protocol to test heavy metals in school water

• Collected, prepared, and analyzed water samples using ICP-MS instrument

• Identified plumbing deficiencies and proposed solutions based on sampling and analytical results

Organic Chemistry Intern MOTE Marine Laboratory, Sarasota, FL June 2006-August 2006

• Tested for the presence of red tide bio-toxin using LC-MS instrument

• Correlated environmental toxin concentrations to human physiological effects

• Conducted routine chemical monitoring in bay and estuary ecosystems

Research Assistant May 2005-July 2008

Toxicology Consultants & Assessment Specialists Inc., Skaneateles, NY

• Synthesized residential and occupational exposure data from legal and medical records

• Conducted literature searches and presented relevant findings

PERSONAL SCHOLARSHIPS, GRANTS, AND AWARDS

Johns Hopkins NIOSH Education and Research Center pilot project award 2016

Student Travel Award, Urban Food Systems Symposium 2016

Johns Hopkins NIOSH Education and Research Center trainee award 2014-2016

Finalist, Abell Award in Urban Policy 2016

Harvard NIOSH Education and Research Center trainee award 2008-2010

136

PROFESSIONAL ORGANIZATIONS, CERTIFICATIONS & CONTINUING EDUCATION CLASSES

AgriSafe Network:

Agricultural Medicine: Occupational and Environmental Health for Rural Health Professionals 2013

FEMA Emergency Management Institute: Radiological Accident Assessment Concepts 2013

National Radon Proficiency Program: Residential Measurement & Mitigation Provider 2012-2015

OSHA: Hazardous Waste Operations and Emergency Response (HAZWOPER) 2010-2015

Professional Association of Diving Instructors (PADI): Advanced Open Water SCUBA Diver 2007

Boy Scouts of America: Eagle Scout 2004

SERVICE & LEADERSHIP

Volunteer Mentor & Head of Family

“Thread” Incentive Mentoring Program, Baltimore, MD 2015-Present

• Led volunteer “family” providing an at-risk youth with academic, social, and employment support

Environmental Health Sciences Representative and Finance Committee Member

JHU Bloomberg School of Public Health Student Assembly, Baltimore, MD 2015-2016

• Represented the Department of Environmental Health Sciences in school governance

• Evaluated applications and awarded funding for student organization programming

Student Assembly Representative

JHU Environmental Health Sciences Student Organization, Baltimore, MD 2015-2016

• Liaised between the Department of Environmental Health Sciences and Student Assembly

Volunteer Bicycle Mechanic

Bike Recycle Vermont, Burlington, VT 2011-2014

• Refurbished bicycles, empowering low-income Vermonters through transportation independence

• Trained clients and other volunteers on bicycle repair

SELECTED PUBLICATIONS AND PRESENTATIONS

Thomas, G., Sousan, S., Tatum, M., Liu, X., Zuidema, C., Fitzpatrick, M., Koehler, K. & Peters, T.

(2018). Low-Cost, Distributed Environmental Monitors for Factory Worker Health. Sensors 18(5):

1411.

Zuidema, C., Tatum, M., Afshar-Mohajer, N., Thomas, G., Peters, T. & Koehler, K. (in revision).

Efficacy of Paired Electrochemical Sensors for Measuring Ozone Concentrations. Submitted to

Journal of Occupational and Environmental Hygiene.

Zuidema, C., Sousan, S., Stebounova, L.V., Gray, A., Liu, X., Tatum, M., Stroh, O., Thomas, G.,

Peters, T. & Koehler, K. (in revision). Mapping Occupational Hazards with a Multi-Sensor Network in

a Heavy-Vehicle Manufacturing Facility. Submitted to Annals of Work Exposure and Health.

137

Afshar-Mohajer, N., Zuidema, C., Sousan S., Hallett, L., Tatum, M., Rule, A.M., Thomas, G., Peters,

T. & Koehler, K. (2018). Evaluation of Low-Cost Electro-Chemical Sensors for Environmental

Monitoring of Ozone, Nitrogen Dioxide and Carbon Monoxide. Journal of Occupational and

Environmental Hygiene 15(2): 87-98.

Zuidema, C., Sousan, S., Stebounova, L., Fitzpatrick, M., Tatum, M., Thomas, G., Peters, T. &

Koehler, K. (2017, October 19). Estimating Personal Exposure to Particulate Matter Using a Low-

Cost Wireless Sensor Network and Indoor Positioning Data. Oral Presentation at the 36th Annual

Conference of the American Association for Aerosol Research (AAAR), Raleigh, NC.

Zuidema, C. (2017, June 13). Using Low-Cost Sensors for Environmental Health Measurements:

Challenges and Opportunities. Annual Meeting of the American Welding Society, Baltimore, MD.

Zuidema, C., Synk, C., Kim, B.F., Harding, J., Rak, S., Emery, M., & Nachman, K.E. (2016, June 23).

Development of a Human Health Risk Assessment Framework for Consumption of Foraged Items in

the Urban Environment: A Baltimore, MD Case Study. Oral Presentation at the Urban Food System

Symposium (UFSS), Olathe, KS.

Zuidema, C., Plate, R. & Dikou, A. (2011). To preserve or to develop? East Bay dredging project,

South Caicos, Turks and Caicos Islands. Journal of Coastal Conservation 15: 555-563.

PROFESSIONAL SKILLS

Languages: English (native speaker); Spanish (intermediate)

Software: Microsoft Office Suite (proficient); MATLAB (proficient); STATA (proficient); ORACLE

Crystal Ball (proficient); ArcGIS (intermediate); R (intermediate)

Other: Sampling environmental media; Industrial Hygiene observational assessments; Data collection,

management, analysis, and visualization; Science communication and education for the general public;

Program management

Estimating Occupational Exposures with a Multi-Hazard ...

Documents