Citation
Analysis of physiological metrics across construction activities

Material Information

Title:
Analysis of physiological metrics across construction activities
Creator:
Majumder, Sreemala Das
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Master's ( Master of science)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Civil Engineering, CU Denver
Degree Disciplines:
Civil engineering
Committee Chair:
Clevenger, Caroline
Committee Members:
Abdallah, Moatassem
Brothers, Heidi

Notes

Abstract:
Construction industry is associated with and responsible for safety management and occupational health of construction workers. Several study shows how physical activity impacts the health and safety of the construction worker at a construction site. However, few studies exist where the impact of construction work on worker health is monitored, taking into account the individual’s physical conditions. This research focuses on collecting physiological and environmental data to explore the impact of construction activity on the individual worker under ambient conditions. The study was conducted with ten US Air Force Academy cadets performing four different construction activities during a summer field course in Colorado Springs, Colorado. A monitoring device was used to record the cadets’ vital signs and physical indicators including heart rate, breathing rate, core temperature, physiological load, mechanical load and posture. The dataset was collected from the cadets performing similar construction activities over a three week period, following similar food and sleep regiments. Results suggest that heat index and construction activities affect the physiological metrics of the individual cadets differently, despite individuals having relatively similar physical characteristics. Furthermore, the results suggest that concrete and asphalt placement are generally the most physically demanding construction activities studied, followed by, heavy equipment and surveying activities respectively. Finally, the study shows that it is possible to independently compare discrete physiological metrics across individuals as well as activities. The research serves to highlight significant opportunity to use such methods to study construction worker health and productivity in the future.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
Copyright Sreemala Das Majumder. Permission granted to University of Colorado Denver to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item has the following downloads:


Full Text
ANALYSIS OF PHYSIOLOGICAL METRICS ACROSS CONSTRUCTION ACTIVITIES
by
SREEMALA DAS MAJUMDER
Diploma,West Bengal State Council of Technical Education,2012 B.Tech., Maulana Abul Kalam Azad University of Technology,2015
A thesis submitted to the Faculty of the Graduate School of University of Colorado in partial fulfillment Of the requirements for the degree of Master in Science Civil Engineering Program 2018
l


This thesis for the Master of Science degree by Sreemala Das Majumder has been approved for the Civil Engineering Program by
Caroline Clevenger, Chair Moatassem Abdallah Heidi Brothers
Date: July 28, 2018
n


ACKNOWLEDGEMENTS
I wish to express my sincerest gratitude to my advisor Dr. Caroline Clevenger for guiding and mentoring me. I have profoundly benefited from her guidance over the past one year. Her enthusiasm for pursuing interesting construction management problems at the highest level of scientific integrity and rigor has been a constant source of inspiration for my research. I deeply thank her for allowing me the space to think myself and for fostering my capacity as a student.
I would like to thank my co-advisor Dr. Moatasssem Abdallah for his support and critical remarks on my work.
Life as an international graduate student is not easy. Mine was no exception. I am fortunate enough to be surrounded by a couple of wonderful friends, who deserves special mention Avijit Shee, Nandana Das- they were beside me in some of my worst time. I could reach out to them any time I needed.
I would like to thank my parents for their faith in me and allowing me to be as ambitious as I wanted. It was under their watchful eye that I gained so much drive and an ability to tackle challenges in life. I would not be here without their sacrifices, patience, continuous support and unconditional love. I would also like to express my gratitude for my in-laws. Dr. Anindita Bandyopadhyay, my mother-in-law, who even after remaining thousands of miles away, has always tried her best to provide me company and support.
Last but not the least, I express my heartiest thanks to my husband Dr. Pratyaydipta Rudra, who has provided immense support in tough times. Without his constant encouragement and love, this journey would not have been possible.
ill


Finally, I dedicate this thesis to my husband for making me who I am, for providing me with
unconditional support and encouragement all throughout. This accomplishment would not have been possible without him.
IV


ABSTRACT
Construction industry is associated with and responsible for safety management and occupational health of construction workers. Several study shows how physical activity impacts the health and safety of the construction worker at a construction site. However, few studies exist where the impact of construction work on worker health is monitored, taking into account the individual’s physical conditions. This research focuses on collecting physiological and environmental data to explore the impact of construction activity on the individual worker under ambient conditions. The study was conducted with ten US Air Force Academy cadets performing four different construction activities during a summer field course in Colorado Springs, Colorado. A monitoring device was used to record the cadets’ vital signs and physical indicators including heart rate, breathing rate, core temperature, physiological load, mechanical load and posture. The dataset was collected from the cadets performing similar construction activities over a three week period, following similar food and sleep regiments.
Results suggest that heat index and construction activities affect the physiological metrics of the individual cadets differently, despite individuals having relatively similar physical characteristics. Furthermore, the results suggest that concrete and asphalt placement are generally the most physically demanding construction activities studied, followed by, heavy equipment and surveying activities respectively. Finally, the study shows that it is possible to independently compare discrete physiological metrics across individuals as well as activities. The research serves to highlight significant opportunity to use such methods to study construction worker health and productivity in the future.
v


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION..................................................1
Literature Review............................................2
II. METHODOLOGY AND RESEARCH.....................................12
Equipment...................................................12
Different Components of the Bioharness Device...............13
Metrics.....................................................13
Data Collection.............................................15
Assumptions.................................................17
Methods.....................................................18
Data filtration and Organization.........................18
Weather Data.............................................20
Model fitting...............................................23
Description of Results......................................27
III. RESULTS......................................................30
Heart Rate................................................30
Breathing Rate............................................38
Posture...................................................46
Core Temperature..........................................56
Physiological Load........................................64
Mechanical Load...........................................72
vi


Activity Level..................................................80
Illustration of the effects using four chosen volunteers........87
IV. CONCLUSION................................................94
Summary and Discussion..........................................94
Delimitations...................................................97
Future Research.................................................97
Bibliography...........................................................99


LIST OF TABLES
TABLE
1. Pilot data metrics.......................................................14
2. Construction Activities..................................................17
3. Characteristics of Individual Participants...............................18
4. List of summary measures to collapse data for different metrics..........20
5. Timetable of Weather Conditions and Construction Activity by Volunteer..22
6. Effect of Gender, Heat Index, Activity Level and Activity on Heart Rate..31
7. Comparison between different activities for Heart Rate...................36
8. Effect of Gender, Heat Index, Activity Level and Activity on Breathing Rate.. 40
9. Comparison between different activities for Breathing Rate...............44
10. Effect of Gender, Heat Index, Activity Level and Activity on Posture....48
n. Comparison between different activities for Posture......................53
12. Effect of Gender, Heat Index, Activity Level and Activity on Core Temperature .............................................................................57
13. Comparison between different activities for Core Temperature............61
14. Effect of Gender, Heat Index, Activity Level and Activity on Physiological Load .............................................................................65
15. Comparison between different activities for Physiological Load..........69
16. Effect of Gender, Heat Index, Activity Level and Activity on Mechanical Load 73
17. Comparison between different activities for Mechanical Load.............77
18. Effect of Gender, Heat Index and Activity on Activity Level.............80
19. Comparison between different activities for Activity Level..............84
V111


20. Comparison between different activities for Heart Rate...............90
21. Comparison between different activities for Breathing Rate...........90
22. Comparison between different activities for Core Temperature.........91
23. Comparison between different activities for Posture..................91
24. Comparison between different activities for Activity Level...........91
25. Comparison between different activities for Mechanical Load..........92
26. Comparison between different activities for Physiological Load.......92
27. Summary Table of all metrics and their association (p-values) with the
explanatory variables..................................................93
IX


LIST OF FIGURES
FIGURE
1. Correlation Heat-Map between different metrics................................25
2. Violin plot for Heart Rate of all volunteers..................................31
3. Violin plot showing the distribution of overall Heart Rate for Female and Male
volunteers....................................................................32
4. Line diagram of aggregated Heart Rate for each value of Heat Index..........33
5. Line diagram of aggregated Heart Rate for each value of Heat Index by
Activities....................................................................34
6. Scatter plot showing the association between Heart Rate and Activity Level.. 35
7. Violin plot showing the distribution of Heart Rate across the four activities... 37
8. Heart rate versus Activity by volunteers, width of each line represents the
Activity Level................................................................38
9. Violin plot for Breathing Rate of all volunteers..............................39
10. Violin plot showing the distribution of overall Breathing Rate for Female and
Male volunteers...............................................................41
n. Line diagram of aggregated Breathing Rate for each value of Heat Index.....42
12. Line diagram of aggregated Breathing Rate for each value of Heat Index by
Activities....................................................................42
13. Scatter plot showing the association between Breathing Rate and Activity
Level.........................................................................43
14. Violin plot showing the distribution of Breathing Rate across the four activities.
..............................................................................45
x


15. Breathing rate versus Activity by volunteers, width of each line represents the
Activity Level...............................................................46
16. Violin plot for Posture of all volunteers...................................47
17. Violin plot showing the distribution of overall Posture for Female and Male
volunteers...................................................................49
18. Line diagram of aggregated Posture for each value of Heat Index.......50
19. Line diagram of aggregated Posture for each value of Heat Index by Activities. ...............................................................................51
20. Scatter plot showing the association between Posture and Activity Level....52
21. Violin plot showing the distribution of Posture across the four activities.54
22. Posture versus Activity by volunteers, width of each line represents the Activity
Level......................................................................55
23. Violin plot for Core Temperature of all volunteers........................56
24. Violin plot showing the distribution of overall Core Temperature for Female and
Male volunteers............................................................58
25. Line diagram of aggregated Core Temperature for each value of Heat Index.. 59
26. Line diagram of aggregated Core Temperature for each value of Heat Index by
Activities.........................................................................59
27. Scatter plot showing the association between Core Temperature and Activity
Level..............................................................................60
28. Violin plot showing the distribution of Core Temperature across the four
activities.........................................................................62
xi


29. Core Temperature versus Activity by volunteers, width of each line represents
the Activity Level...........................................................63
30. Violin plot for Physiological Load of all volunteers........................64
31. Violin plot showing the distribution of Physiological Load for Female and Male
volunteers...................................................................66
32. Line diagram of aggregated Physiological Load in log scale for each value of
Heat Index...................................................................67
33. Line diagram of aggregated Physiological Load for each value of Heat Index.. 67
34. Scatter plot showing the association between Physiological Load and Activity
Level........................................................................68
35. Scatter plot showing the association between Physiological Load and Activity
Level........................................................................70
36. Physiological Load versus Activity by volunteers, width of each line represents
the Activity Level...........................................................71
37. Violin plot for Mechanical Load of all volunteers...........................72
38. Violin plot showing the distribution of Mechanical Load for Female and Male
volunteers...................................................................74
39. Line diagram of aggregated Mechanical Load in log scale for each value of Heat
Index........................................................................75
40. Line diagram of aggregated Mechanical Load for each value of Heat Index____75
41. Scatter plot showing the association between Mechanical Load and Activity
Level
76


42. Violin plot showing the distribution of Mechanical Load across the four activities.....................................................................
78
43. Mechanical Load versus Activity by volunteers, width of each line represents
the Activity Level...............................................................79
44. Violin plot for Activity Level in log scale of all volunteers...................80
45. Violin plot showing the distribution of overall Activity Level in log scale for
Female and Male volunteers.......................................................82
46. Line diagram of aggregated Activity Level in log scale for each value of Heat
Index............................................................................83
47. Line diagram of aggregated Activity Level in log scale for each value of Heat
Index by Activities..............................................................84
48. Violin plot showing the distribution of Activity Level in log scale across the four
activities.......................................................................85
49. Violin plot showing the distribution of Heart Rate for volunteer 1,2,3,4 across
the four activities..............................................................86
50. Violin plot showing the distribution of Breathing Rate for volunteer 1,2,3,4
across the four activities.......................................................86
51. Violin plot showing the distribution of Core Temperature for volunteer 1,2,3,4
across the four activities.......................................................86
52. Violin plot showing the distribution of Posture for volunteer 1,2,3,4 across the
four activities (taking absolute value for Posture)..............................87
53. Violin plot showing the distribution of Posture for volunteer 1,2,3,4 across the
four activities.
87


54. Violin plot showing the distribution of Activity Level for volunteer 1,2,3,4
across the four activities.........................................................87
55. Violin plot showing the distribution of Mechanical Load for volunteer 1,2,3,4 across the four activities............................................................88
56. Violin plot showing the distribution of Physiological Load for volunteer 1,2,3,4
across the four activities.
88


CHAPTER 1: INTRODUCTION
Construction is a labor intensive industry where construction workers are subjected to intensive physical stress and strain, awkward work postures, heavy weight lifting, and a range of work postures (Hartmann and Fleischer 2005) .Injuries from construction activity can be immediate or cumulative over time. This can lead to decreases in work productivity, inattentiveness or inability to make wise decisions. Demanding workload can take a toll on both mental and physical health of a worker. Furthermore, continuous depression and dissatisfaction from work-site can result in unwanted accidents and more injuries (Abdelhamid and Everett 2002). Construction workers are exposed to ergonomic hazards, which includes dynamic movement, various awkward postures, pulling and lifting loads (Hartmann and Fleischer 2005; Schneider and Susi 1994). They are at risk to develop musculoskeletal disorders (Forde et al. 2005; Goldsheyder et al. 2002; Griffin et al. 2000; Holmstrom and Engholm 2003; Reinhold-Keller et al. 2000; Sandmark 2000; Schneider 2001). However, previous studies have primarily been limited to one or more physical indicators of the construction worker. For example, research by (Wang and Fu 2016) and (Kirk and Sullman 2001) studied heart rate and demonstrated that heart rate can be used to detect physiological strain in different situations. The physical characteristics such as age, height and weight also impact the stress and strain the individual experiences.
1


Literature Review:
In this section, the author will discuss previous research focused on occupational health hazards and the safety issues of construction workers at jobsite, as well as explore which factors have been shown to influence health and safety on a construction site. Many means for detecting these factors exist. This literature review specifically considers the collection of live data from construction workers and its analysis to find association between the physiological metrics and environmental factors.
Many construction activities require extensive physical labor over long period of time in adverse weather conditions. Construction activities often include heavy lifting, pulling and pushing and various awkward postures. (Hartmann and Fleischer 2005) suggests that construction work, safety and productivity are connected with each other and can influence individual construction worker health. Measuring physical strain can show how the worker’s occupational health is affected, and, consequently, help to maintain and promote safety on the job site (Abdelhamid and Everett 2002; Bouchard and Trudeau 2008; Garet et al. 2005).
Correlating the health of a construction worker with physiological measures to detect physical strain is a relatively new concept. One of the first studies in this area was done by (Kirk and Sullman 2001). They used heart rate as the main indicator to determine physical strain of cable hauler choker setters in New Zealand. Results identified a correlation of heart rate to physical strain and overall assessment of workload for a specific task. This study questioned the ability to collect accurate real time data when the workers are in motion without causing personal discomfort. A
2


similar study by (Abdelhamid and Everett 2002) added oxygen consumption to heart rate data to determine the performance of 100 construction workers doing moderate to heavy work. The data showed that the average oxygen uptake was 0.82 L/ min (± 0.22 L/min) and the average heart beat was 108 beats/min (± 17 beat) min). These measurements were compared to standard guidelines for acceptable levels of physical performance for specific industrial environment. The results indicated that 20-40% of the workers’ regularly exceed threshold levels determined by standard published guidelines for manual work. Exceeding the threshold limits cause workers to be more prone towards inattentiveness, decreased productivity, poor judgment, accidents and injuries in job site.
(Roja et al. 2006) conducted a study on workers from the heavy civil industry which included ten road construction and maintenance workers and ten pavers who belonged from the age group 20-60 years. The study measured the physical demands of road construction work and estimated muscle fatigue. Average metabolic energy consumption for road construction and repairing works were recorded as 8.1± 1.5 kcal/min, and7.2 ± 1.1 kcal/min for paving. Their findings, based on average metabolic energy consumption, indicated that road construction work requires extreme manual labor, compulsive working posture, and continuous arm and leg movements. In the same study they monitored the workers’ heart rate, posture and muscle tone for a week and suggested that work related musculoskeletal problems could be common for these workers. (Taylor et al. 2010) claimed that the symptoms of musculoskeletal disorders (MSD) are common for young construction workers. The chance of MSD increase with the increasing years and women are more effected than men by MSD. Early prevention
3


strategies can reduce disability associated with MSD. Another similar study was conducted on highway tunnel construction workers by (Tak et al. 2011) focusing primarily on “PATH” (Posture, Activity, Tools and Handling) to determine health hazards. They found that non-neutral trunk postures and, in some operations, kneeling and squatting are fairly common. These postures pose risk for musculoskeletal disorders, effecting the back and knees in particular.
Construction workers in some part of the world are exposed to extreme heat. Climatic Heat stress can be fatal. (Rowlinson et al. 2014) studied factors affecting climatic heat stress and identified three ways of reducing heat stress in construction sites: 1) Control of climatic heat stress exposure through use of an action-triggering threshold system, 2) control of Continuous Work Time with mandatory work-rest routine, and 3) allow workers to follow self-pace regimes. However, each construction site is unique and each job site should have its own set of heat stress guidelines.
(Bates and Schneider 2008) conducted a study on construction workers in thermally extreme conditions in United Arab Emirates (UAE) and monitored temperature, fluid intake, urine specific gravity and heart rate. They concluded that people working in extreme conditions can avoid fatigue when provided with plenty of fluid and are allowed to self-pace work. Other research studied the impact of the climate in Hong Kong on construction workers. In Hong Kong, the industry demands long and irregular hours. (Yi and Chan 2015) established a Monte Carlo simulation-based method to find the optimal work pattern. To sustain the climatic adversity the authors proposed a schedule with a direct work rate of 87.8% of the entire work time as an ideal work pattern for the construction workers in Hong Kong. Job rotation was also
4


considered to be an effective strategy to promote health in these working conditions. Finally, consistent, period breaks are recommended to minimize occupational health hazards and maximize productivity.
Another strategy to mitigate climatic heat stress as suggested by (Yi et al. 2016) is to use an early-warning system that prompts signals to worker in distress. This system is developed with the help of smart bracelet and a smart phone that collects data from the construction worker. Artificial neural Network is used to obtain the final signals. The heat strain here is calculated from a subjective index rating. The proposed system can be used to protect the well-being of the workers’ as the new algorithm helps to send out early warning alerts to the construction workers.
A strong connection exists between physical performance and hydration. Dehydration leads to changes in cardiovascular, thermoregulatory, metabolic and central nervous system function that hinders the performance, and heat exposure along with dehydration can be fatal situation for construction workers on the job. (Murray and Murray 2017) studied the relationship between hydration and physical performance. Practical guidelines at construction site encourages consumption of water without overdrinking. Reducing sweating rate, taking frequent rest break, drinking sufficient volumes of fluid are the simplest ways to improve physical performance.
Recently, significant research has focused on the roll technology can play in monitoring worker health and safety. (Cheng and Teizer 2013) studied the opportunity to use video cameras on construction job sites to track worker dynamics.
5


(Marx and Konig 2011) introduced physical monitoring system (PMS) to estimate construction workers’ physical strain based on heart rate and acceleration and found out two of the three PMS system are reliable based on different activities and testing. The PMS devices selected for this study are Zephyr BioHarness (BH)- BT, Zephyr HxM, and Hidalgo EQ-01. (Hwang and Lee 2017) have shown the association between Heart Rate Reserve percentages (%HRR) and physiological demand. This study revealed physical demand is dependent largely on changing working conditions, demand of the task, age, working conditions and working patterns. (Romagnol 2010) showed how real time data analysis can very effective and efficient during emergency and extreme condition of construction workers. (Buller et al. 2010) discussed the importance of estimation of core temperature of human body during construction activity. Another research used real time location sensing (RTLS) and physiological status monitoring (PSM) to monitor the posture and physical status of the workers. (Cheng et al. 2013) (Hwang and Lee 2017) verified the importance of continuous physical measurement to detect any significant and sudden change in their physical demand.
Smart mobile phones, medical sensors, wireless communication system, portable trackers and health monitoring systems have been used to track continuous real time data for medical patients, workers and athletes. (Banos et al. 2014) discussed a PhysioDroid that monitors data remotely based on vital signs like electrocardiogram, heart rate, respiration rate, skin temperature, and body motions. The versatility of the device is a key factor. Monitoring construction workers’ physiological conditions, however, can differ from sports, science and medicine. (Gatti et al. 2014) investigated the validation of PMSs for construction based on two physiological parameters such as
6


heart rate and breathing rate. Results indicated that PMS can successfully analyze association between physical strain and task level, productivity assessment and safe-unsafe behavior of construction workers. (Scheer et al. 2014) collected heart rate data, weather data and video recordings to establish the relation that PSM technology certainly indicates overexertion, potential fatigue, illness, and cardiovascular overload.
Another study involving PSM and heart rate data tried to measure the physiological cost of concrete construction activities (Lee and Migliaccio 2016). Heart rate was observed to be a reliable predictor of physiological stress level. Physiological cost of construction activities were calculated from HR indexes and descriptive statistics, including mean, standard deviations, and modified Karvonen’s formula. Physiological cost comparison revealed concrete construction activities to be moderate and comparable to steelwork. Placing and vibrating concrete produced more physical strain, and requires more onsite personal safety measures. However, the study showed that physiological cost varies over time. The higher cost are observed during the midday and at the beginning and end of each week.
Another study on roof-top construction workers’ was conducted to estimate Total Worker Health (TWH) by (Lee et al. 2017). TWH merge occupational health hazards with the workers’ off duty wellbeing. The study focused roofing workers since they frequently encounter high on duty health and safety risks and poor-off duty lifestyles. The study also validated the reliability and usability of wearable sensors.
(Wang et al. 2017) proposed the use of a wireless and wearable electroencephalography (EEG) system to process the workers’ brain signals. Results from their on-site experiment show EEG signals such as frequency, power spectrum
7


density and spatial distribution can effectively detect the construction workers’ perceived level of risk.
(Zhou and Ding 2017) proposed an Internet-of-Things (loT) - based safety barrier alert system to avoid hazardous injury for underground construction. This technology has been implemented in the construction of the Yangtze River- crossing metro station to test its performance. The system generates timely early warning alerts under critical environment using spatial-temporal barrier. (Aryal et al. 2017) incorporated a heart rate monitor, infrared temperature sensors device and an ECG sensor enabled live data monitoring for 12 individuals. Their results suggest that data combined from all the sensors were 82% accurate in predicting physical fatigue. (Awolusi et al. 2018; Yu et al. 2017) also validated the use of real time data based for physiological monitoring, environmental sensing, proximity detection and location tracking analysis on construction workers
Worker injuries and accidents, equipment damage directly contributes to disrupted work schedules, worker compensation, insurance hike, financial loss. Advanced safety efforts in construction site can avoid or minimize all these costs. (Abudayyeh et al. 2006) concluded how the management committee and safety managers working as a team can improve the overall conditions by focusing on engineering improvements along with emphasizing human behavior commitment from the entire organization.
Worker training programs can improve safety performance and enhance work efficiency. (Teizer et al. 2013) in their study proposed a new approach to integrate
8


real time data from construction job site and incorporate them in prior education and training programs. With real time location tracing and visualization technology, they helped workers to identify safety issue that they were not aware of due to their natural working environment and personal habits. Combination of real time data tracking and visualization technology was shown to be advantageous in terms of improving occupational health hazards.
Construction workers are frequently required to work at high heights and managing fall from heights (FFH) is one of the essential part of construction safety management. Numbers show that FFH is one of the leading cause for serious injuries and fatalities across the world. (Guo and Goh 2017) identified the causes of FFH and develop a universal ontology model that would benefit all construction workers from building and construction industries. (Fang and Dzeng 2017) designed a fall detecting method based on a hierarchical threshold-based algorithm to identify and avoid fall accidents. (Yang et al. 2017) proposed a sensing device approach that understands the workers’ gait irregularities to identify physical fall in a construction site. In this study, the wearable sensory devices used to collect kinematic gait data were attached to the workers’ ankle. The cumulative gait data was analyzed to obtain gait abnormality score representing the hazardous situations in job site.
(Choi et al. 2017) investigated how to motive construction workers to put on wearable devices, in this case, smart vest with GPS device on them and a wristband-type wearable activity tracker, and found that workers were more likely to wear such devices if they were aware of the potential benefits. General Behavior Based Safety (BBS) is based on visual recording and manual observation, which is an impractical
9


solution for the construction jobsite. However, (Yu et al. 2017) suggested a theoretical framework using an image skeleton- based parameterized method. This research improves the effectiveness of the construction safety management.
(Choe and Leite 2017) formulated a 4 dimensional (4D) construction safety method which is specific to space, time, and location. Visual safety materials were used to enhance safety communication at jobsite. A particular case study revealed that risky activity, days and zones can be distinguished from safer activity, days and zones. This research helps the safety personnel with a proactive, job site specific safety planning model that helps to manage the construction site.
Finally, (Hwang and Lee 2017) successfully converted the heart rate to heart rate reserve (%HHR) to set a threshold on acceptable amount of time a construction activity needs a specific amount of % HRR. Once the % HRR is proved to be consistent, it can be used in different construction equipment to measure heart rate as % HRR. Hwang and Lee (2017) established successful relationships between % HRR and influencing factors to show how the wristband is feasible to capture workers’ any significant high physical demand during construction work.
Despite significant research into construction worker health and safety, and recent focus on using technology to assess spatial and physical conditions on construction sites, a critical need remains for monitoring and collecting physiological data of construction workers. This research addresses this need by collecting physiological and environmental data on US Air Force Academy cadets performing four different construction activities during a summer field course in Colorado Springs,
10


Colorado. The author using the data set to explore and statistically compare the impact of construction activity on individual workers under similar environmental conditions across activities.
11


CHAPTER 2: METHODOLOGY AND RESEARCH
Equipment:
Various sensors have been used and validated in the past to monitor physical activities. Physiological status monitors (PSMs) is a readily available and effective device for monitoring heart rate, breathing rate and acceleration (Zhen Wang & Shan Fu, 2016). However, heart rate is significantly related to breathing rate and body acceleration, and is also effected by an individual’s age, physical fitness, daily routine, food habits and other personal factors. While heart rate can detect fatigue, with more indicators there will be better chance to assess performance ability. Research has shown that environmental condition like temperature and humidity also impact fatigue.
For this study, the author used Zypher Bioharness system to monitor volunteers.
Zephyr Technologies specializes in remote physiological monitoring and location tracer without hindering the flexibility and freedom of the any individual. This technology has been used to collect data from professional sports person, special force, rescue workers, medical field and in other research studies. The Zephyr Bio-harness system are most commonly used in sports and sport medicine to optimize the work-rest ratio, allowing for a better heart rate recovery regime and optimized physical performance. ( Zephyr 2012, 2013, 2016)
12


Different Components of the Bioharness Device:
The following Zephyr Bio-harness components were used for this study.
1. Bio-harness Strap- an approximately one inch wide strap worn across the chest with the puck inserted in the desired slot. There is a compression strap which allows the puck to be on exact position against the sternum. The correct position of the puck is along the edge of the strap under the arm-pit, or slightly to the rear.
2. Puck- the main sensory device that is pressed by the compression strap against the sternum to collect data.
3. Puck docking station- holding device that charges the pucks and downloads the data.
4. OmniSense Software- a proprietary software which allows viewing of live data and recorded data. It also includes capabilities to analyze the data using predefined algorithms.
Metrics:
This research applied the following metrics: - heart rate, breathing rate, core temperature, mechanical load, physiological load and posture. The following are the description are based on and extend those provided by the manufacturer (Zephyr 2013).
13


Table 1: Pilot data metrics
Metric Description Units Notes
Heart Rate It is measured as the number of heart beat per minute. Heart beats per minute The measure of Heart rate is analyzed from the 250Hz Echocardiogram (ECG) data
Breathing Rate It is measured as the number of breathes per minute. Breathing per minute The sensors inside the zephyr puck detects breathing by the expansion and contraction of our torso.
Posture It is measured as the change of angle of any individual in comparison with the gravity. Degree from vertical position. When any individual is standing straight the measurement is zero. Forward and backward leaning accounts for positive and negative values.
Core Temperature The core temperature is calculated based on the heart rate, skin temperature and a predefine formula by Zephyr system Degree Centigrade The accuracy of this estimate and have also demonstrated that such a computational measurement can indicate physical stress before an individual reaches an unhealthy state (Buller and Hoyt 2008).
Mechanical Load Mechanical intensity over time, where mechanical intensity is a measure of instantaneous effort based on acceleration Unitless Measured based on the movement of an individual and acceleration.
Physiological Load Physiological intensity over time, where physiological intensity is a measure of instantaneous effort based on heart-rate Unitless Measured based on cardiovascular output.
Activity Level The activity metric is a collective measurement of the accelerometer given in velocity magnitude units (VMU) ranging between Og to 16g. Acceleratio n due to gravity The averages of the three axial acceleration magnitudes over the previous 1 second epoch, sampled at 100Hz.
14


Data Collection:
This study was conducted on 10 cadets from the 2017 summer program at the Field Engineering and Readiness Laboratory (FERL) at the United States Air Force Academy. The cadets self-volunteered for this study. International Review Board protocols for research involving human subjects at both University of Colorado Denver and US Air Force Academy (USAFA) were completed and thoroughly complied to. FERL has a 3-week long regime in June. USAFA’s FERL handbook describes the program as, “cadets to actual hands-on experiences in surveying, construction methods, and construction materials... using a ‘construct first, design later’ approach” ("Civil Engineering Practices - Field Engineering Cadet Handbook," 2017). The program is designed such that all cadets are exposed to classroom and field activities that give them an experience about civil construction activities. The cadets were performing all the activities at a high elevation. We did not consider any additional effect of high altitude on the physiological metrics of the cadets. The cadets are grouped in different “flights” or groups of twelve individuals. The study was conducted over three weeks where the volunteer subjects were exposed to various construction work in regard to civil engineering like concrete placement, surveying etc. It is important to note that all the volunteers were strictly following similar food and sleep regiments while staying at the academy. The schedules for each flight was different from one another. However the schedule is designed such that all flights had to perform all activities over the course of three week. For our study we mainly focused on 4 activities: concrete placement,
15


heavy equipment, surveying and asphalt paving. For simplification of our analysis we numbered the different activities which are included in Table 2.
The data collected from the ten volunteers comprised of 8 male volunteers and 2 female volunteers. All the cadets were within the age range of 20-22. Prior to the study we collected the individual characteristics of the cadets which included weight, height, and age. This data was imported by the software “Omnisense” to analyze the physiological status for each volunteer (Zephyr 2016) to the software also requires input of a fitness level for each individual. Physiological status monitoring is a way to collect and record the vital signs of the volunteers. It is a nonintrusive method which can occur in real time. Since all of the volunteers were cadets in the United States Air force and they were in their early twenties, each was assigned an eight out of ten for fitness.
Data were collected over three weeks. However there were some missing data when volunteers forgot to switch on the device. Moreover there were some inconsistencies when the straps were loose and did not fit the cadets properly. To maintain the quality of the data, they were filtered and treated before running the analysis. Omnisense software generates a metric called Heart Rate Confidence (HRC). HRC (%) is calculated based on the electrocardiogram (ECG), ECG noise and worn detection. The threshold for accepting data was based on HRC of 80% or more, based on recommendation of Zephyr’s representative.
16


Assumptions:
1. For our research we assumed that all our cadet volunteers (aged between 20-22 years) have a high level of fitness. The United States Air force Academy at Colorado Springs is situated at an elevation of 7258 feet. All the activities were performed in this high elevation.
2. They volunteers were from different flights performing similar activities over the three week summer program. We assumed all the volunteers were doing exact similar activities on different days.
3. We assumed that the Zephyr Bioharness Pucks were pre calibrated.
Table 2: Construction Activities
Activity Activity Number Description
Concrete Placement (CP) 1 Prepare a site, set framework and reinforcement, place concrete, and take samples and perform slump test
Heavy Equipment (HE) 2 Operate construction equipment including an excavator, scraper, bulldozer, loader, and paving machine.
Surveying (S) 3 Use total station to measure distance and horizontal/vertical angles. Lay out a calculated location for concrete slab pour and use measuring techniques to plot land data.
Asphalt Paving (AP) 4 Place a section of road using approximately 20 tons of asphalt.
17


Table 3: Characteristics of Individual Participants
Volunteer Gender Height (feet, inches) Weight (pounds) Age (years)
1 M 5'11" 154 22
2 M 6'2" 187 21
3 M 6'0" 183 21
4 M 6'0" 170 20
5 F 5'11" 175 21
6 F 5'2" 145 20
7 M 5'11" 171 20
8 M 57" 170 21
9 M 6'4" 180 20
10 M 6'4" 180 22
Methods:
Data filtration and organization:
The Bioharness puck system has sufficient memory to store 36 hours of data at one minute interval. The data can be viewed either in real time using a blue-tooth connection to OmniSense Live software, or can be viewed after downloading the OmniSense Analysis software. For our study, the data for all the ten volunteers at one second interval were downloaded each day manually.
Statistical software R was used to run all the statistical analysis. Omnisense software allows user to download excel files for further study and research purpose which is beyond the scope of Omnisense software. These excel files can be imported using R and subsequently analyzed.
Our research focused particularly on four activities 1. Concrete 2.Heavy equipment 3. Surveying 4. Asphalt. The first step is to filter the day we need the data
18


for our analysis, since there were other activities that the volunteers performed over the three weeks of summer program. Once the days were selected, all the data were concatenated to form a single excel file and purposely named as the “Main Table”. This data frame was used to run the analysis for our research.
The analysis part comprises of two major sections. The first part includes all the filtration and organization while the other part is mostly about fitting models to answer our research questions. The steps for the organization part are as follows.
Step 1: The data frame was filtered by HRC %>. Any row in the data matrix with HRC below 80 was removed.
Step 2: Only the columns corresponding to the metrics that we are interested in (e.g., heart rate, breathing rate etc.) were kept and the rest were removed.
Step 3: The data set contains observations at one second intervals. For the sake of simplification and convenience we collapsed the data for every second to data for every minute by taking a summary measure within each minute. Depending on the type of metric different summary measures were decided for different metrics. For most of our metrics, average was used to collapse data within each minute. However, mechanical and physiological loads are cumulative load that builds up over time. Thus the gain for each minute (maximum-minimum) was calculated for these two metrics. Table 4 shows the different summary measures. Finally, before fitting any model or further analysis the data set was checked for any abnormal values or outliers.
19


Table 4: List of summary measures to collapse data for different metrics
Metric Summary measure to collapse data within every minute
Heart Rate Average for each minute.
Breathing Rate Average for each minute.
Posture Average for each minute.
Core Temperature Average for each minute.
Mechanical Load Gain for each minute.
Physiological Load Gain for each minute.
Activity Level Average for each minute.
Weather Data:
Weather parameters influence the physiological metrics for any construction worker. Previous studies have indicated that exposure to the extreme temperature can influence the physical strain of a construction worker working outdoors. We collected temperature and humidity data for the time period when the volunteers were performing their activities. The source for our data was United States Air Force Academy weather station as reported on the Weather Underground website (https://www.wunderground.com/history).
After collecting the temperature and relative humidity data, the researchers evaluated the Heat Index (HI) based on Ersatz version of the heat index equation (Rothfusz 1979) given by
Heat Index = -42.379 + 2.04901523 x T + 10.14333127 x R - 0.22475541 xTxR-6.83783 x 10“3 x T2 - 5.481717 x 10“2 x R2 + 1.22874 x 10“3 x T2 x R + 8.5282 x 10-4 x T x R2 - 1.99 x 10-6 x T x 2 x R Equation 1
Where, T=ambient dry bulb temperature (°F)
20


R= Relative Humidity (integer Percentage)
The activity schedule was divided into morning and afternoon shifts and two values of Heat Index were available for each day.
Table 5 shows the days and the heat index values for morning and evening session.
21


Table 5: Timetable of Weather Conditions and Construction Activity by Volunteer
DAYS TEMPERATU RELATIVE HEAT CONCRETE HEAVY SURVEYING ASPHALT
RE (°F) HUMIDITY INDEX EQUIPMENT
(%)
6/5/2017 65.3 52.83 78.32 V-6,10
DAY 1 74.4 28.2 76.61 F-B
6/6/2017 64.6 49.5 78.99 V-8
DAY 2 64.7 56 78.30 F-C
6/7/2017 64.3 58.83 78.09 V-9,5
DAY 3 70.1 47.8 77.02 F-D
6/9/2017 75.4 23.17 76.48 V-1,2,3,4
DAY 5 85.1 12.2 81.93 F-A
6/12/2017 69.9 54 76.64 V-9,5
DAY 6 83.3 13.8 80.58 F-D
6/13/2017 65.9 19.5 73.22 V-1,2,3,4
DAY 7 76.6 6.4 75.28 F-A
6/14/2017 68 20 73.75 V-1,2,3,4 V-6,10
DAY 8 79 13.4 77.59 F-A F-B
6/15/2017 71.2 23.83 75.28 V-6,10 V-8
DAY 9 83.8 11.4 80.94 F-B F-C
6/16/2017 72 28.5 76.09 V-8 V-9,5
DAY 10 87.1 10.8 83.52 F-C F-D
6/196/201 67.9 38.33 77.26 V-6,10
7 80.8 25.2 79.36 F-B
DAY 11
6/20/2017 73.2 37.83 77.00 V-8
DAY 12 88.3 15 84.52 F-C
6/21/2017 80.3 29.83 79.35 V-9,5
DAY 13 87.1 17 83.58 F-D
6/23/2017 50.6 82.17 88.13 V-1,2,3,4
DAY 15 F-A
22


Model fitting:
Multiple Linear regression model is predominantly used to model relationship among two or more explanatory variables and a response variable by fitting a linear equation to the observed data. The equation for a multiple linear regression with p explanatory variables has the form
Yi= po + Pi*ii + P2Xi2 + PsXis + "â– PpXip + Cl, (Equation 2)
where, Yt= response/dependent variable for ith sample (]k = Estimated regression coefficients,
xik = value of the explanatory/independent variables for ith sample, k = 1,2, ...,p.
€i= error (residual)
In case of a multiple linear regression it is assumed that all et are independent.
However, for our data there are repeated observations over time for an individuals which cannot be assumed to be independent. An appropriate way to model such repeated measures data across time is the use of linear mixed effects model.
The Linear Mixed Effect Model (LMM) comprises of both fixed effects and random effects. The random effects take care of the correlation between the repeated observations. For our study we aimed to figure out the effect of the independent variables such as gender, heat index, activity level and activity on the metrics. The LMM is represented as follows.
23


(Equation 3)
Yij = Po + Pl^ij + P2Xijl + Pl^ijl + P3Xij3 ^-E Pp^ijp + bi + €lj>
where, Ytj = Value of the response for the )th replication for ith individual bt= individual specific random effect for ith individual Xj = explanatory/independent variables, i = 1,2, ...,p.
€ij= error (residual)
The model assumes
eirN( 0,ae2),
bi-NiQ.al),
and they are independent.
Specifically for our data, the model can be written as
Yij = Po + Pigender + p2H.I + @3 A.L + p4Activity2 + p5Activity3 + p6Activity4 +
bi + eU (Equation 4)
Ytj = Value of the response for the )th replication for ith individual
Activity2- if vo^unteer does activity 2
0 otherwise
Activjty3- v°lunteer does activity 3
0 otherwise
Acyvit 4- if v°lunteer does activity 4 t 0 otherwise


r__,, r 1 Male
G lO Female
H.l= Heat Index
A.L= Activity Level
bj= individual specific random effect for ith individual
6ij= error (residual)
0.1278 0.4096 0.6403 1
0.0795 0.3453 1 0.6403
0.2895 1 0.3453 0.4096
1 0.2895 0.0795 0.1278
Heart Rate Breathing Rate Posture Core Temperature Physiological Load Mechanical Load Activity Level
value
1.00
0.75
0.50
0.25
0.00
Figure 1: Correlation Heat-Map between different metrics.
Prior to deciding about our model, we computed the correlations among the different metrics to understand their interrelations as well as to decide which variables to include in the model. Figure 1 shows the correlation between all the metrics for our study. This heat-map was generated by coloring according to the value of the Pearson’s correlation coefficient. The color bar in the figure shows the scale for strong and weak correlations.
25


Although Activity Level is one of our metrics of interest, we also include it in the models (except when the response is Activity Level) as an explanatory variable since it has high correlation with other variables and can be thought of as a causal variable that can influence the responses. It can influence other metrics but not the other way around. Since Activity Level has a skewed distribution, we use log transformed values when considering it as the response variable in our models.
The LMM is used to answer most of our research questions. The research questions and the corresponding null hypotheses are as follows.
Q1. For each metric, is the outcome statistically similar for all activities?
H0:(34 = 0 &/?5 = 0 &/?6 = 0 vs HA: H0 is not true.
Q2. For each metric, is there any statistical difference in the outcome between men and women?
H0:px = 0 vs Ha\/3i =£ 0
Q3. For each metric, is the outcome statistically similar across different heat Index? Ho-.Pi = 0 vs HA\ (32^0
Q4. For each metric, is the outcome statistically similar across different activity levels? Ho-.Ps = 0 vs Ha: /?3 =£ 0
The hypotheses can be tested using statistical test of significance. T-tests are used to test the hypotheses for Q2, Q3 and Q4, while a Likelihood Ratio Test (LRT) is used to test the hypothesis for Q1. If a significant difference between different
26


activities is found for Q1, we then followed up using pairwise comparisons (t-test) to test which activities have significantly different outcomes.
Since multiple hypotheses are tested, we need to adjust the raw p-values using methods for multiple testing adjustment. We used Bonferroni correction (Dunn 1961) for this purpose.
The metrics Physiological Load and Mechanical Load are observed to be not only positively skewed, but also zero inflated. Therefore, regular LMM is not appropriate for these metrics. Instead, we use Compound Poisson Generalized Mixed Model (CPGLMM), a model specially designed for such data (Zhang 2014).
Description of results
The following is a list of the visualization techniques and a summary of the statistical data presented for each metric in the Results Section to follow. Results are presented using violin plots, as well as tables listing T-test p-values and pairwise comparison using Bonferroni adjusted p-values. For each metric the following information is provided:
1. Violin Plot- Violin plots visualize the distribution of the data. They are vertically symmetric presentations of the probability distribution of the response variable as estimated from the data. The mean and the mean ± 1 standard deviation range are also shown. The following sections present the results for each metric in succession.
2. Table showing statistical analysis of effect of Gender, Heat Index, Activity Level and Activity on Metrics. The following generic table explains in details the representative symbols along with their meaning and description. To test
27


for statistical differences a T-test was performed for the metric data across all volunteers. Table shows the fitted LMM and the p-values corresponding to the hypotheses to be tested for Q1, Q2, Q3 and Q4. Column 1 represents the effects and column 4 represents the estimates of the corresponding regression coefficients (/3's). The test statistics and the Bonferroni-adjusted p-values are reported in column 5 and 6 respectively.
Table: Description of statistical data analyzed and presented for each metric.
Effect Coefficient Interpretation Estimat e Test statistic P value
Gender Pi Average difference in HR between males and females.
Heat Index P2 Average increase in the HR when HI increases by one unit.
Activity level P3 Average increase in the HR when AL increases by one unit.
Activity P4 Average difference in HR for activities HE & C “
Ps Average difference in HR for activities S & C “
Pe Average difference in HR for activities A & C “
Note: Fit of LMM for Heat Rate showing the effect of Gender, Heat Index, Activity Level and Activity on Heart Rate. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0.
3. Line diagrams -A line diagram is a graph that connects a series of points by drawing line segments between them. These points are ordered in one of
28


their coordinate (either the x-coordinate or y-coordinate) value. Line diagrams are commonly used to represent trends in data.
4. Scatter plots- A graph in which the values of two or more variables are plotted along two axes, the pattern of the resulting points reveals if there is any correlation present in the data set.
5. Pairwise Statistical Comparison table- This table consist of pairwise follow up comparison between different activities. Each cell in the table shows Bonferroni adjusted p-value for the corresponding pairwise comparison. The pairwise comparisons of p-values indicates statistically distinction in metric profiles among different activities.
Table: Comparison between different activities for Metric
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - - - -
Heavy Equipment
Surveying - - - -
Asphalt - - - -
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
29


CHAPTER 3: RESULTS
Results
The following results address the research questions. The following sections present the results for each metric in succession.
Heart Rate:
Table 6 shows the Heart rate distributions for the 10 volunteers. Different profiles are evident for different Volunteers. As shown, some Volunteer have higher mean Heart rates than others.
To test for statistical differences a t-test was performed for the Heart Rate data across all volunteers as shown in Table 7.
30


160
1 23456789 10
Volunteer
Figure 2: Violin plot for Heart Rate of all volunteers.
Table 6: Effect of Gender, Heat Index, Activity Level and Activity on Heart Rate
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. 1.074 t=0.177 1
Heat Index P2 Average increase in the HR when HI increases by one unit. -0.096 t=-3.491 0.033
Activity level P3 Average increase in the HR when AL increases by one unit. 103.081 t=81.397 0
Activity P4 Average difference in HR for activities HE & C -1.472 /2=309. 31 6.6234 x 10-65
Ps Average difference in HR for activities S & C -3.841
Pe Average difference in HR for activities A & C -0.594
Notes: Fit of LMM for Heat Rate showing the effect of Gender, Heat Index, Activity Level and Activity on Heart Rate. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0.
31


The p-values indicate that Heat Index, Activity Level and Activity are significantly associated with Heart Rate (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 3 shows the frequency distribution of Heart Rate for females and males. It visually confirms the results of Table 6 that Heart Rate differences are not statistically significant across gender.
Figure 3: Violin plot showing the distribution of overall Heart Rate for Female and Male volunteers.
32


Figure 4 and Figure 5 depict the relationship of Heart Rate to Heat Index with each colored line representing an individual volunteer. Although the p-values indicate significant associations significant, it is difficult to draw clear conclusions about patterns in the relationship between Heart Rate and Heat Index. For example, /?2 = -0.0964 implies that with a change of Heat Index from its minimum observed value 73.22 to its maximum observed value 88.36, the Heart Rate decreases only by 1.46, on average. Since the number of observations is very large, the p-values can be very small even when there is a small, practically negligible effect size. While a general trend may be that Heart Rate declines as Heat Index rises (as most clearly demonstrated by V1-V4), more research is recommended to further analyze this relationship.
o
o
75
80
85
Heat Index
Figure 4: Line diagram of aggregated Heart Rate for each value of Heat Index.
33


Heart Rate Heart Rate
Concrete
Heavy Equipment
t
ro
a>
I
Heat Index
Heat Index
Surveying
Asphalt
Figure 5: Line diagram of aggregated Heart Rate for each value of Heat Index by Activities.
Figure 6 plots Heart rate vs. Activity level, colored by Activity and suggests a positive correlation (upward trend) exists between Heart Rate and Activity Level.
34


Activity Level
Figure 6: Scatter plot showing the association between Heart Rate and Activity Level.
Figure 6 plots Heart rate vs. Activity level, colored by Activity and suggests a positive correlation exists between Heart Rate and Activity Level.
Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Heart Rates. Results are summarized in Table 7.
35


Table 7: Comparison between different activities for Heart Rate
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - 4.07E-08 0 1
Heavy Equipment 4.07E-08 0 0.301
Surveying 0 0 - 0
Asphalt 1 0.301 0 -
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
The pairwise comparisons of p-values indicate that Concrete-Heavy Equipment, Concrete-Surveying, Heavy Equipment-Surveying, Heavy Equipment-Asphalt result in statistically distinction Heart Rate profiles from one another. Finally, the Heart Rate during surveying is the most distinct compared to the other three Activities. Figure 7 shows that Heart Rate is, on average, lowest during surveying and spikes highest during Concrete Placement.
36


160-
120-
a>
-t—'
03
a:
â– t:
TO
CD
X
80-
40-
1
Activities
Concrete
Heavy Equipment Surveying Asphalt
2 3 4
Activity
Figure 7: Violin plot showing the distribution of Heart Rate across the four activities.
To understand the relationship of Heart Rate with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Heart rate for each volunteer with the width of the lines showing the amount of Activity Level in Figure 8. As expected, the lines become wider towards the top of the figure, but that is not always the case. For example, for volunteer 4, we clearly see the pattern Concrete > Heavy Equipment > Surveying > Asphalt, despite higher activity levels with Asphalt compared to Surveying. Other similar patterns are
37


observed, but there is no consistent pattern across all volunteers. This implies that there is interaction between volunteer and Activity and including that in the model might be ideal. However, since Activities are modeled by fixed effects and volunteers are modeled by random effects, it can only be achieved using a highly complicated model which is beyond the scope of this thesis. However, there is no consistent pattern across all volunteers, which implies more research is recommended to analyze these relationships.
fl if?
Vc Activity lunteer 8
X f 1 ! ~ * t T — i
â–  Concrete
â–  Heavy Equipment H Surveying
â–¡ Asphalt
Figure 8: Heart rate versus Activity by volunteers, width of each line represents the Activity Level.
Breathing Rate:
Figure 9 shows the Breathing Rate distributions for the 10 volunteers. Different Volunteers have different Breathing Rate profiles. As shown, some Volunteer have higher mean Breathing Rates than others.
38


Volunteer
Figure 9: Violin plot for Breathing Rate of all volunteers.
To test for statistical differences a t-test was performed for the Breathing Rate data across all volunteers. Table 7 shows the results of the statistical testing.
The p-values indicate that Heat Index, Activity Level and Activity are significantly associated with Breathing Rate (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 10 - Figure 13 show the associations between Breathing Rate and the independent variables. Figure 10 shows the distribution of Breathing Rate by Gender.
It visually confirms the results of Table 8 that Breathing Rate differences are not statistically significant across gender.
39


Table 8: Effect of Gender, Heat Index, Activity Level and Activity on Breathing Rate
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. -0.666 t=- 0.753 1
Heat Index P2 Average increase in the HR when HI increases by one unit. 0.030 t=3.773 0.011
Activity level P3 Average increase in the HR when AL increases by one unit. 16.238 t=44.11 7 0
Activity P4 Average difference in HR for activities HE & C -0.026 /2=70.5 5 2.2450 x 10“13
Ps Average difference in HR for activities S & C -0.483
Pe Average difference in HR for activities A & C -1.033
Notes: Fit of LMM for Breathing Rate showing the effect of Gender, Heat Index, Activity Level and Activity on Breathing Rate. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0.
40


0-
Female
Male
Gender
Figure 10: Violin plot showing the distribution of overall Breathing Rate for Female and Male volunteers.
Figure 11 and Figure 12 depict the relationship of Breathing Rate to Heat Index with each colored line representing an individual volunteer. Although the p-values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Breathing Rate and Heat Index. While a general trend may be that Breathing Rate declines as Heat Index rises, more research is recommended to further analyze this relationship.
41


Q)
CO
DU
O
C\l
CO
CO
i-----------------------------1-----------------------------r
75 80 85
Heat Index
Figure 11: Line diagram of aggregated Breathing Rate for each value of Heat Index.
Concrete Heavy Equipment
Figure 12: Line diagram of aggregated Breathing Rate for each value of Heat Index by Activities.
42


Breathing Rate
Figure 13 plots Breathing Rate vs. Activity level, colored by Activity and suggests a positive correlation (upward trend) exists between Breathing Rate and Activity Level.
o
co
LO
CM
O
CM
ID
O
LO
O
â–¡ Concrete
â–¡ Heavy Equipment
â–¡ Surveying
â–¡ Asphalt
0.0 0.2 0.4
0.6
Activity Level
Figure 13: Scatter plot showing the association between Breathing Rate and Activity Level.
43


Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Breathing Rates. Results are summarized in Table 9.
Table 9: Comparison between different activities for Breathing Rate
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - 1 1.18 X KT11 1
Heavy Equipment 1 5.67 x 1(T10 1
Surveying 1.18 x lO-11 5.67 x 1(T10 - 0.0002
Asphalt 1 1 0.0002 -
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
The pairwise comparisons of p-values indicate that Concrete-Surveying, Heavy Equipment-Surveying, Surveying-Asphalt result in statistically distinction Breathing Rate profiles from one another. Finally, the Breathing Rate during surveying is the most distinct compared to the other three Activities.
Figure 14 visually confirms that Breathing Rate differs by Activity with Surveying having the lowest average Breathing Rate and Asphalt having the maximum range of Breathing Rate. However, it is difficult to discern which activities are statistically distinct.
44


30-
CD
ro
a:
o>
£Z
20
cc
CD
m
10-
Activities
Concrete Heavy Equipment Surveying Asphalt
0-
12 3 4
Activity
Figure 14: Violin plot showing the distribution of Breathing Rate across the four activities.
Finally, to further explore the relationship of Breathing Rate with the Activities also accounting for the difference between volunteers and the difference in Activity Level, Figure 15 is a plot of Activity versus Breathing Rate for each volunteer with the width of the lines showing the amount of Activity Level. This figure reveals how the Breathing Rate varies based on the choice of Activity for different volunteers. As expected, the lines become wider towards the top of the figure, but that is not always the case. For example, for volunteer 10, we see the pattern Asphalt > Surveying >Heavy Equipment, despite having similar Activity Levels. However, there is no consistent
45


pattern across all volunteers, which implies more research is recommended to analyze these relationships.
H
Activity Volunteer 6
l \ {i i t +
l - ^ '11
Activity Volunteer 8
ill it
B Concrete B Heavy Equipment B Surveying â–¡ Asphalt
Figure 15: Breathing rate versus Activity by volunteers, width of each line represents the Activity Level.
Posture:
Figure 16 shows the Posture distributions for the 10 volunteers. Different profiles are in Posture based are evident for different Volunteers. As shown, some Volunteers have higher mean Posture than others.
46


100-
-100 -
i 23456789 10
Volunteer
Figure 16: Violin plot for Posture of all volunteers.
To test for statistical differences a T-test was performed for the Posture data across all volunteers as shown in Table 10.
The p-values indicate that Heat Index, Activity Level and Activity are significantly associated with Posture (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 17 - Figure 20 show the associations between Posture and the independent variables.
Figure 17 shows the distribution of Posture by Gender. It visually confirms the results of Table 10 that Posture differences are not statistically significant across gender.
47


Table 10: Effect of Gender, Heat Index, Activity Level and Activity on Posture
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. 3.657 0.770 1
Heat Index P2 Average increase in the HR when HI increases by one unit. -0.043 -0.806 1
Activity level P3 Average increase in the HR when AL increases by one unit. 14.158 5.664 1.04 x 10“6
Activity P4 Average difference in HR for activities HE & C -5.330 /2=408. 565 2.2 x 10“16
Ps Average difference in HR for activities S & C 3.274
Pe Average difference in HR for activities A & C 2.523
Notes: Fit of LMM for Posture showing the effect of Gender, Heat Index, Activity Level and Activity on Posture. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0.
48


Posture
100-
50-
o-
-50-
-100
Female
Gender
Male
Figure 17: Violin plot showing the distribution of overall Posture for Female and Male volunteers.
49


o
co
Heat Index
Figure 18 and Figure 19 depict the relationship of Posture to Heat Index with each colored line representing an individual volunteer. Although the p-values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Posture and Heat Index. While a general trend may be that Posture declines as Heat Index rises this is not, necessarily intuitive, and more research is recommended to further analyze the relationship.
50


Posture
O _
co
Heat Index
Figure 18: Line diagram of aggregated Posture for each value of Heat Index.


Posture Posture
Concrete
Heavy Equipment
Surveying
Asphalt
Figure 19: Line diagram of aggregated Posture for each value of Heat Index by Activities.
Figure 20 plots Posture vs. Activity level, colored by Activity and suggests a positive correlation exists between Posture and Activity Level.
52


Posture
o
o
o
LO
• •
o -
o
in
• •
o
o
r.:
â–¡ Concrete
â–¡ Heavy Equipment
â–¡ Surveying
â–¡ Asphalt
T
0.0
0.2
0.4
0.6
Activity Level
Figure 20: Scatter plot showing the association between Posture and Activity Level.
Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Posture. Results are summarized in Table 11.
53


Table 11: Comparison between different activities for Posture
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - 0 1.29 X KT11 0
Heavy Equipment 0 0 0
Surveying 1.29 X HT11 0 1
Asphalt 0 0 1
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
The pairwise comparisons of p-values indicate that Concrete- Heavy Equipment, Concrete-Surveying, Concrete- Asphalt, Heavy Equipment-Surveying, Surveying-Asphalt result in statistically distinction Posture profiles from one another. Finally, the Posture during Heavy Equipment is the most distinct compared to the other three Activities.
Figure 21 visually confirms that Posture differs by Activity with Heavy Equipment having the lowest average Posture. However, it is difficult to discern which activities are statistically distinct.
54


100-
Activity
Activities
Concrete
Heavy Equipment Surveying Asphalt
Figure 21: Violin plot showing the distribution of Posture across the four activities.
To understand the relationship of Posture with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Posture for each volunteer with the width of the lines showing the amount of Activity Level (Figure 22). This figure reveals how the Posture varies based on the choice of Activity for different volunteers. As expected, the lines become wider towards the top of the figure, but that is not always the case. For example, for
55


volunteer 8, we see how the average and range of Posture is much larger as compared to Heavy equipment and Surveying despite similar Activity Levels. However, there is no consistent pattern across all volunteers. This implies that there is interaction between Volunteer and Activity and including that in the model might be ideal, but that can only be achieved using a highly complicated model which is beyond the scope of this thesis.
Volunteer 1
Volunteer 2
Volunteer 3
Volunteer 4
Volunteer 5
. J i * 0 20 40 60 80 100 iff" 0 20 40 60 80 100 I i j i I 1 i I 0 20 40 60 80 100 1 J I | 1 s‘ I i i
Activity Activity Activity Activity Activity
Volunteer 6 Volunteer 7 Volunteer 8 Volunteer 9 Volunteer 10
2 ' 2 ' 2 ' 2 ‘
3 § — fi o _ 5? o _ 3 Q- o . ~ ^ 0- O . Q- O . 0. o -
f I t O - I | | o - i t t â– =r ill
Activity Activity Activity Activity Activity
s Concrete â–  Heavy Equipment B Surveying â–¡ Asphalt
Figure 22: Posture versus Activity by volunteers, width of each line represents the Activity Level.
Core Temperature:
Figure 23 shows the Core Temperature distributions for the 10 volunteers. Different Volunteers have different Core Temperature profiles. As shown, some Volunteer have higher mean Core Temperatures than others.
56


39-
(D
Z3
2
ai
CL
E
£
CD
o
O
38-
37-
36-
35-
123456789 10
Volunteer
Figure 23: Violin plot for Core Temperature of all volunteers.
To test for statistical differences a t-test was performed for the Core Temperature data across all volunteers as shown Table 12.
The p-values indicate that Heat Index, Activity Level and Activity are significantly associated with Core Temperature (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 24 - Figure 27 show the associations between Core Temperature and the independent variables.
Figure 24 shows the distribution of Core Temperature by Gender. It visually confirms the results of Table 12 that Core Temperature differences are not statistically significant across gender.
57


Table 12: Effect of Gender, Heat Index, Activity Level and Activity on Core
Temperature
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. 0.095 t=0.943 1
Heat Index P2 Average increase in the HR when HI increases by one unit. 0.004 t=5.777 5.32 x 10“7
Activity level P3 Average increase in the HR when AL increases by one unit. 0.420 t=12.531 0
Activity P4 Average difference in HR for activities HE & C 0.0178 X2=1660.6 0
Ps Average difference in HR for activities S & C -0.191
Pe Average difference in HR for activities A & C -0.163
Notes: Fit of LMM for Core Temperature showing the effect of Gender, Heat Index, Activity Level and Activity on Core Temperature. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0.
58


39-
CD
38-
CT3
i_
0)
Q.
E
CD
© 37-
o
O
36-
35-
Female
Male
Gender
Figure 24: Violin plot showing the distribution of overall Core Temperature for Female and Male volunteers.
Figure 25 and Figure 26 depict the relationship of Core Temperature to Heat Index with each colored line representing an individual volunteer. Although the p-values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Core Temperature and Heat Index. While a general trend may be that Core Temperature declines as Heat Index rises, it is not the case for all volunteers. More research is recommended to further analyze this relationship.
59


Heat Index
Figure 25: Line diagram of aggregated Core Temperature for each value of Heat Index.
Concrete
Heavy Equipment
CL
E
o
O
77 78 79 80
Heat Index
81
82
CL
E
o
O
74
76
78 80
Heat Index
82
a.
E
o
O
74
Surveying
76
78
80
82
Heat Index
CL
E
o
O
78
80
Asphalt
82 84
Heat Index
86
88
Figure 26: Line diagram of aggregated Core Temperature for each value of Heat Index by Activities.


Figure 27 plots Core Temperature vs. Activity level, colored by Activity and suggests a positive correlation exists between Core Temperature and Activity Level.
Activity Level
Figure 27: Scatter plot showing the association between Core Temperature and Activity Level.
Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Core Temperature. Results are summarized in Table 13.
61


Table 13: Comparison between different activities for Core Temperature
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - 0.324 0 0
Heavy Equipment 0.324 0 0
Surveying 0 0 - 0.032
Asphalt 0 0 0.032 -
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
The pairwise comparisons of p-values indicate that Concrete-Surveying, Concrete-Asphalt, Heavy Equipment-Surveying, Heavy Equipment-Asphalt, Surveying-Asphalt result in statistically distinction Heart Rate profiles from one another. Finally, the Core Temperature during Surveying and Asphalt is the most distinct.
Figure 28 visually confirms that Core Temperature differs by Activity with Surveying having the lowest average Core Temperature.
62


39-
Figure 28: Violin plot showing the distribution of Core Temperature across the four activities.
To understand the relationship of Core Temperature with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Core Temperature for each volunteer with the width of the lines showing the amount of Activity Level (Figure 29). This figure reveals how the Core Temperature varies based on the choice of Activity for different volunteers. We observe some difference in average Core temperature as well as the range of Core Temperatures based on activities but there is no consistent pattern across all
63


volunteers. This implies that there is interaction between volunteer and Activity and including that in the model might be ideal, but that will require a highly complicated model which implies more research is recommended to analyze these relationships.
Volunteer 1
o
£ n s
I o

b ” T
o °
<~l «
o
n Activity
Volunteer 6
o
§ « “
2 o 5 i T

b « - T J
i p p
Activity

Volunteer 2
Activity Volunteer 7
e
8
.
Volunteer 3
Volunteer 4
Activity
Volunteer 5
Activity
B Concrete B Heavy Equipment o Surveying â–¡ Asphalt
Figure 29: Core Temperature versus Activity by volunteers, width of each line represents the Activity Level.
Physiological Load:
Figure 30 shows the Physiological Load distributions for the 10 volunteers. Different Volunteers have different Physiological Load profiles. As shown, some Volunteer have higher mean Physiological Loads than others. The maximum buildup of Physiological Load for Volunteer 1, 3 and 8 is higher compared to the others.
64


â– O
03
o
4-
2
0-





( 1 1 » 1 < t t > —< , i < i , ) *— * k

2 4 6 o"
Volunteer
Figure 30: Violin plot for Physiological Load of all volunteers.
To test for statistical differences t-test was performed for the Physiological Load data across all volunteers as shown Table 14.
The p-values indicate that Heat Index, Activity Level and Activity are significantly associated with Physiological Load (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 31 - Figure 37 show the associations between Physiological Load and the independent variables. Figure 31 shows the distribution of Physiological Load by Gender.
65


Table 14: Effect of Gender, Heat Index, Activity Level and Activity on Physiological Load
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. 0.947 X2=1.077 1
Heat Index P2 Average increase in the HR when HI increases by one unit. -0.039 X2=28.603 6.13 x 10“6
Activity level P3 Average increase in the HR when AL increases by one unit. 11.808 X2=2108.35 0
Activity P4 Average difference in HR for activities HE & C -0.812 X2=591.631 4.53 x 10-126
Ps Average difference in HR for activities S & C -1.359
Pe Average difference in HR for activities A & C -0.966
Notes: Fit of CPGLMM for Physiological Load showing the effect of Gender, Heat Index, Activity Level and Activity on Physiological Load. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0. All p-values are conducted using likelihood ratio test.
66


6-
â– o
Ct3
5 4-
03
O
O)
_o
O
to
>*
Female Male
Gender
Figure 31: Violin plot showing the distribution of Physiological Load for Female and Male volunteers.
Figure 32 and Figure 33 depict the relationship of Physiological Load to Heat Index with each colored line representing an individual volunteer. Although the p-values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Physiological Load and Heat Index. While a general trend may be that Physiological Load declines as Heat Index rises, more research is recommended to further analyze this relationship.
67


Heat Index
Figure 34: Line diagram of aggregated Physiological Load in log scale for each value of Heat Index.
Concrete Heavy Equipment
Surveying
Asphalt
Heat Index
Heat Index
Figure 35: Line diagram of aggregated Physiological Load for each value of Heat Index.


Physiological Load
Figure 36 plots Physiological Load vs. Activity level, colored by Activity and suggests a positive correlation exists between Physiological Load and Activity Level.
Activity Level
Figure 36: Scatter plot showing the association between Physiological Load and Activity Level.
69


Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Physiological Loads. Results are summarized in Table 15.
Table 15: Comparison between different activities for Physiological Load
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - 0 0.0032 2.76 X 10“13
Heavy Equipment 0 0 0.054
Surveying 0.0032 0 - 0
Asphalt 2.76 X 10-13 0.054 0.003709 -
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
The pairwise comparisons p-values indicate that all activities result in statistically distinction Physiological Load profiles from one another.
Figure 37 visually confirms that Physiological Load differs by Activity. However, it is difficult to discern which activities are statistically distinct.
70


6-
Activities
Concrete Heavy Equipment Surveying Asphalt
Figure 37: Scatter plot showing the association between Physiological Load and Activity Level.
To understand the relationship of Physiological Load with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Physiological Load for each volunteer with the width of the lines showing the amount of Activity Level (Figure 38). This figure reveals how the Physiological Load varies based on the choice of Activity for different volunteers. The overall Physiological Load build up for activity Concrete dominates other activities (Figure 38), which is also visible in Figure 37. This implies that there is interaction
71


between volunteer and Activity and including that in the model might be ideal, but that will require a highly complicated model which implies more research is recommended to analyze these relationships.
Physiological Load also has a lot of zeros in the data which restricts us to obtain dependable results and conclusion.
Volunteer 1 Volunteer 2 Volunteer 3 Volunteer 4 Volunteer 5
ii, Physiological Load 0 1 2 3 4 5 6 7 lii. •C M ■ a o - 1 i * t Physiological Load 0 1 2 3 4 5 6 7 * - Physiological Load 0 1 2 3 4 5 6 7 S it
Activity Activity Activity Activity Activity
Volunteer 6 Volunteer 7 Volunteer 8 Volunteer 9 Volunteer 10
h' ‘ — r- ‘
-o “ ‘ ra ^ -o ‘ ra ^
O m - o - — o U) - o in -
5 -a- - 8 - — . — 8 ■» - 8 -=r -
S’ — S’ S’ S'
t w . â–  = I 5 i *
fit “■
_s_ o — * M. o - f 4 f o - o - £ 3E X
Activity Activity Activity Activity Activity
â–¡ Concrete
â–  Heavy Equipment la! Surveying
â–¡ Asphalt
Figure 38: Physiological Load versus Activity by volunteers, width of each line represents the Activity Level.
Mechanical Load:
Figure 39 shows the Mechanical Load distributions for the 10 volunteers. Different Volunteers have different profiles in Mechanical Load and it is difficult to observe based on the scale of the graph. However, it is possible to observe that some Volunteer have higher mean Mechanical Loads than others.
72


5-
4-
Volunteer
Figure 39: Violin plot for Mechanical Load of all volunteers.
To test for statistical differences a t-test was performed for the Mechanical Load data across all volunteers. Table 16 shows the fitted CPGLMM and the p-values corresponding to the hypotheses to be tested for Q1, Q2, and Q3.
The p-values indicate that Heat Index, Activity Level and Activity are significantly associated with Mechanical Load (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 40 - Figure 43 show the associations between Mechanical Load and the independent variables. Figure 40 shows the distribution of Mechanical Load by Gender.
73


Table 16: Effect of Gender, Heat Index, Activity Level and Activity on Mechanical
Load
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. -0.169 /2=0.472 1
Heat Index P2 Average increase in the HR when HI increases by one unit. 0.007 /2=1.850 1
Activity level P3 Average increase in the HR when AL increases by one unit. 15.401 /2=6184.8 45 0
Activity P4 Average difference in HR for activities HE & C 0.834 /2=544.32 5 8.14 x 10-116
Ps Average difference in HR for activities S & C -0.277
Pe Average difference in HR for activities A & C -0.040
Notes: Fit of CPGLMM for Mechanical Load showing the effect of Gender, Heat Index, Activity Level and Activity on Mechanical Load. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’ means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0. All p-values are conducted using likelihood ratio test.
74


5-
4
"ro 3-
03
o
'c
03
o 2- 1 -
0
Female
Gender
I
Figure 40: Violin plot showing the distribution of Mechanical Load for Female and Male volunteers.
Figure 41 and Figure 42 depict the relationship of Mechanical Load to Heat Index with each colored line representing an individual volunteer. Although the p-values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Mechanical Load and Heat Index. While a general trend may be that Mechanical Load declines as Heat Index rises, more research is recommended to further analyze this relationship.
75


Mechanical Load Mechanical Load ,s- Mechanical Load
Heat Index
|ure 41: Line diagram of aggregated Mechanical Load in log scale for each value of Heat Index.
Concrete
Heavy Equipment
Surveying
Asphalt
Figure 42: Line diagram of aggregated Mechanical Load for each value of Heat Index.


Figure 43 plots Mechanical Load vs. Activity level, colored by Activity and suggests a low positive correlation exists between Heart Rate and Activity Level.
0.0 0.2 0.4 0.6
Activity Level
Figure 43: Scatter plot showing the association between Mechanical Load and Activity Level.
Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Mechanical Load. Results are summarized in Table 17.
77


Table 17: Comparison between different activities for Mechanical Load
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete 2.46 X 10“64 0.0001 1
Heavy Equipment 2.46 X 1(T64 2.41 X 10-" 3 X 10“35
Surveying 0.0001 2.41 x 10~" - 1
Asphalt 1 3 x 10“35 1 -
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
The pairwise comparisons of p-values indicate that Concrete-Heavy Equipment, Concrete-Surveying, Heavy Equipment-Surveying, Heavy Equipment-Asphalt result in statistically distinction Mechanical Load profiles from one another. . Finally, the Mechanical Load during Heavy Equipment is the most distinct compared to the other three Activities.
78


5-
4-
I I I I
12 3 4
Activity
Activities
Concrete
Heavy Equipment Surveying Asphalt
Figure 44: Violin plot showing the distribution of Mechanical Load across the four activities.
Figure 44 visually confirms that Mechanical Load differs by Activity with Surveying having the lowest average Mechanical Load.
To understand the relationship of Mechanical Load with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Mechanical Load for each volunteer with the width of the lines showing the amount of Activity Level (Figure 45). This figure reveals how the Mechanical Load varies based on the choice of Activity for different volunteers.
79


However, due to numerous zero values in the data, we cannot conclude anything else from Mechanical Load. More research is recommended to analyze these relationships.
Volunteer 1 Volunteer 2 Volunteer 3 Volunteer 4 Volunteer 5
fit Mechanical Load 0 1 2 3 4 5 ■ A « ^ Mechanical Load 0 1 2 3 4 5 W W 1 Mechanical Load 0 1 2 3 4 5 J I I I I L « S ; Mechanical Load 0 1 2 3 4 5 ■ i *
Activity Activity Activity Activity Activity
Volunteer 6 Volunteer 7 Volunteer 8 Volunteer 9 Volunteer 10
- 10 ' .o - 10 '
"S
.3 _3
S 3 " 3 3
c c c c
-C - JZ « - j= rM - .c rv -
0) 1) 0) 0)
5 2 2 2
I s o - i t i ; o - W A ¥ o - i , . o - X W X
Activity Activity Activity Activity Activity
la! Concrete
â–  Heavy Equipment
â–  Surveying â–¡ Asphalt
Figure 45: Mechanical Load versus Activity by volunteers, width of each line represents the Activity Level.
Activity Level:
Figure 46 shows the Heart rate distributions for the 10 volunteers. Different Volunteers have different profiles for Activity Level. As shown, some Volunteer have higher mean Activity Level than others. For example volunteer 6 and 7 seemed to have the highest means as compared to the others.
80


-1
-2
>
o
<
-3-
-4-
-5-
Volunteer
Figure 46: Violin plot for Activity Level in log scale of all volunteers.
To test for statistical differences a T-test was performed for the Activity Level
data across all volunteers as shown Table 18.
Table 18: Effect of Gender, Heat Index and Activity on Activity Level
Effect Coefficient Interpretation Estimate Test statistic P value
Gender Pi Average difference in HR between males and females. -0.364 t=-3.374 0.630
Heat Index P2 Average increase in the HR when HI increases by one unit. -0.008 t=-4.068 0.003
Activity P4 Average difference in HR for activities HE & C -0.077 X2=316.46 1.88 x 10-66
Ps Average difference in HR for activities S & C -0.295
Pe Average difference in HR for activities A & C -0.200
Notes: Fit of LMM for Activity Level showing the effect of Gender, Heat Index, Activity Level and Activity on Activity Level. The p-values for Gender. Heat Index and Activity Level are calculated using t-test and the p-value for Activity was calculated using likelihood ratio test. Note that in interpretation column the ‘difference between X and Y’
81


means X-Y and assumes that the values of all other variables remain the same. The test statistic and the p-values are related to the hypothesis that the difference/increase is 0.
The p-values indicate that Heat Index and Activity are significantly associated with Activity Level (p < 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( Figure 47 - Figure 49 show the associations between Activity Level and the independent variables.
Figure 47 shows the distribution of Activity Level by Gender. It visually confirms the results of Table 18 that Activity Level differences are not statistically significant across gender.
82


Female
Gender
Male
Figure 47: Violin plot showing the distribution of overall Activity Level in log scale for Female and Male volunteers.
Figure 48 and Figure 49 depict the relationship of Activity Level to Heat Index with each colored line representing an individual volunteer. Although the p-values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Activity Level and Heat Index. While a general trend may be that Activity Level declines as Heat Index rises, there are a few exceptions, and more research is recommended to further analyze this relationship. Similarly as explained for Heart Rate earlier, the change of Activity Level will be negligible for a unit change in Heat Index.
83


Activity Level
Since the number of observations is very large, the p-values can be very small even when there is a small, practically negligible effect size.
Heat Index
Figure 48: Line diagram of aggregated Activity Level in log scale for each value of Heat Index.
84


Concrete
Heavy Equipment
Surveying
Asphalt
78 80 82 84 86 88
Heat Index
Figure 49: Line diagram of aggregated Activity Level in log scale for each value of Heat Index by Activities.
Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Activity Level. Results are summarized in Table 19.
Table 19: Comparison between different activities for Activity Level
Activity Concrete Heavy Equipment Surveying Asphalt
Concrete - 0.002 0 0
Heavy Equipment 0.002 “ 0 2.97 X 10“5
Surveying 0 0 0.003
Asphalt 0 2.97 X 10“5 0.003
Notes: Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p-value for the corresponding pairwise comparison.
85


Activity Level
The pairwise comparisons p-values indicated that all activities result in
statistically distinction Activity Level profiles from one another.
-1 -
-2
-3-
-4
-5
12 3 4
Activity
Activities
Concrete
Heavy Equipment
Surveying
Asphalt
Figure 50: Violin plot showing the distribution of Activity Level in log scale across the four activities.
Figure 50 visually confirms that Activity Level differs by Activity with Surveying
having the lowest average Activity Level, and Concrete Placement having the highest. However, it is difficult to discern which activities are statistically distinct.
86


Full Text

PAGE 1

i ANALYSIS OF PHYSIOLOGICAL METRICS ACROSS CONSTRUCTION ACTIVITIES by SREEMALA DAS MAJUMDER Diploma,West Bengal State Council of Technical Education,2012 B.Tech., Maulana Abul Kalam Azad University of Technology,2015 A thesis submitted to the Faculty of the Graduate School of University of Colorado in partial fulfillment Of the requirements for the degree of Master in Science Civil Engineering Program 2018

PAGE 2

ii This thesis for the Master of Science degree by Sreemala Das Majumder has been approved for the Civil Engineering Program by Caroline Clevenger, Chair Moatassem Abdallah Heidi Brothers Date: July 28, 2018

PAGE 3

iii ACKNOWLEDGEMENT S I wish to express my sincerest gratitude to my advisor Dr. Caroline Clevenger for guiding and mentoring me. I have profoundly bene fited from her guidance over the past one year. Her enthusiasm for pursu ing interesting construction management p roblems at the highest level of scientifi c integrity and rigor has been a constant source of inspiration for my research. I deeply tha nk her for allowing me the space to th ink myself and for fostering my capacity as a student . I would like to thank my co advisor Dr. Moatasssem Abdallah for his support and critical remarks on my work. Life as an international graduate student is not easy. Mine was no exception. I am fortunate enough to be surrounded by a couple of wonderful friends, who deserves special mention Av ijit Shee, Nandana Das they were beside me in some of my worst time. I could reach out to them any time I needed. I would like to thank my parents for their faith in me and allowi ng me to be as ambitious as I wanted. It was under their watchful eye that I gained so much drive and an ability to tackle challenges in life. I would not be here without their sacrifices, patience, continuous support and unconditional love. I would also like to express my gratitude for my in laws. Dr. Anindita Bandyopadhyay, my mother in law, who even after remaining thousands of miles away, has always tried her best to provide me company and support. Last but not the least, I express my heartiest thanks to my h usband Dr . Pratyaydipta Rudra , who has provided immense sup port in tough times. Without his constant encouragement and love, this journey would not have been possible.

PAGE 4

iv Finally, I dedicate this thesis to my husband for making me who I am , for providing me with unc onditional support and encouragement all throughout . This accomplishment would not have been possible without him.

PAGE 5

v ABSTRACT Construction industry is associated with and responsible for safety management and occupational health of construction worker s . Several study shows how physical activity impact s the health and safety of the construction worker at a construction site. However, few studies exist where the impact of construction work on worker health is monitored , taking into account the indi focuses on collecting physiological and environmental data to explore the impact of construction activity on the individual worker under ambient conditions. The study was conducted with ten US Air Force Academy c adets performing four different construction activities during a summer field course in Colorado Springs, Colorado. A monitoring including heart rate, breathing rate, core temperatur e, physiological load, mechanical load and posture. The dataset was collected from the cadets performing similar construction activities over a three week period, follow ing similar food and sleep regiments . Results suggest that heat index and construction activities affect the physiological metrics of the individual cadets differently, despite individuals having relatively similar physical characteristics. Furthermore, the results suggest that concrete and asphalt placement are generally the most physically demanding construction activities studied , followed by, heavy equipment and surveying activities respectively. Final ly, the study shows that it is possible to independently compare discrete physiological metrics across individuals as well as act ivities. The research serves to highlight significant opportunity to use such methods to study construction worker health and productivity in the future.

PAGE 6

vi TABLE OF CONTENTS CHAPTER I. INTRODUCTION . Literature Review II. METHODOLOGY AND RESEARCH 12 Equipment Different Components of the Bioharness Device . 13 Metrics Data Collection Methods . Data filtration and Organization Weather Data Model fitting Description of Resu III. RESULTS . 30 . 38 Physiological Lo Mechanical Loa d 72

PAGE 7

vii Illustration of the effects using four chosen volunteers IV. CONCLUSION 94 Bibliography 99

PAGE 8

viii LIST OF TABLES TABLE 1. Pilot data metrics ................................ ................................ ........ 14 2. Construction Activities ................................ ................................ .. 17 3. Characteristics of Individual Participants ................................ ............ 18 4. List of summary measures to collapse data for di fferent metrics ................ 20 5. Timetable of Weather Conditions and Construction Activity by Volunteer ..... 22 6. Effect of Gender, Heat Index, Activity Level and Activity on Heart Rate ....... 31 7. Comparison between different activities for Heart Rate .......................... 36 8. Effect of Gender, Heat Index , Activity Level and Activity on Breathing Rate .. 40 9. Comparison between different activities for Breathing Rate ..................... 44 10. Effect of Gender, Heat Index, Activity Level and Activity on Posture ........... 48 11. Comparison between different activities for Posture .............................. 53 12. Ef fect of Gender, Heat Index, Activity Level and Activity on Core Temperature ................................ ................................ ............................. 57 13. Comparison between different activities for Core T emperature ................. 61 14. Effect of Gender, Heat Index, Activity Level and Activity on Physiological Load ................................ ................................ ............................. 65 15. Comparison between different activities for Physiological Load ................. 69 16. Effect of Gender, Heat Index, Activity Level and Activity on Mechanical Load 73 17. Comparison between different activities for Mechanical Load ................... 77 18. Effect of Gender, Heat Index and Activity on Activity Level ...................... 80 19. Comparison between different activities for Activity Level ....................... 84

PAGE 9

ix 20. Comparison between different activities for Heart Rate .......................... 90 21. Comparison between different activities for Breathing Rate ..................... 90 22. Comparison between different activities for Core Temperature ................. 91 23. Comparison between different activities for Posture .............................. 91 24. Comparison between different activities for Activity Level ....................... 91 25. Comparison between different activities for Mechanical Load ................... 92 26. Comparison between different activities for Physiological Load ................. 92 27. Summary Table of all metrics and their association (p values) with the explanatory variables ................................ ................................ ... 93

PAGE 10

x LIST OF FIGURES FIGURE 1. Correlation Heat Map between different metrics. ................................ . 25 2. Violin plot for Heart Rate of all volunteers. ................................ ......... 31 3. Violin plot showing the distribution of overall Heart Rate for Female and Male volunteers. ................................ ................................ ................ 32 4. Line diagram of aggregated Heart Rate for each value of Heat Index. .......... 33 5. Line diagram of aggregated Heart Rate for each value of Heat Index by Activities. ................................ ................................ ................. 34 6. Scatter plot showing the association bet ween Heart Rate and Activity Level. . 35 7. Violin plot showing the distribution of Heart Rate across the four activities. .. 37 8. Heart rate versus Activity by volunteers, width of each line represents the Activity Level. ................................ ................................ ............ 38 9. Violin plot for Breathing Rate of all volunteers. ................................ .... 39 10. Violin plot showing the distribution of overall Breathing Rate for Female and Male volunteers. ................................ ................................ ......... 41 11. Line diagram of aggregated Breathing Rate for each value of Heat Index. ..... 42 12. Line diagram of aggregated Breathing Rate for each value of Heat I ndex by Activities. ................................ ................................ ................. 42 13. Scatter plot showing the association between Breathing Rate and Activity Level. ................................ ................................ ...................... 43 14. Violin plot showing the distribution of Breathing Rate across the four activities. ................................ ................................ ............................. 45

PAGE 11

xi 15. Breathing rate versus Activity by volunteers, width of each line represents the Activity Level. ................................ ................................ ............ 46 16. Violin plot for Posture of all volunteers. ................................ ............. 47 17. Violin plot showing the distribution of overall Post ure for Female and Male volunteers. ................................ ................................ ................ 49 18. Line diagram of aggregated Posture for each value of Heat Index. .............. 50 19. Line diagram of aggregated Posture for each value of Heat Index by Activities. ................................ ................................ ............................. 51 20. Scatter plot showing the association between Posture and Activity Level. ..... 52 21. Violin plot showing the distribution of Posture across the four activities. ...... 54 22. Posture versus Activity by volunteers, width of each line represents the Activity Level. ................................ ................................ ...................... 55 23. Violin plot for Core Temperature of all volunteers. ................................ 56 24. Violin plot showing the distribution of overall Core Temperature for Female and Male volunteers. ................................ ................................ ......... 58 25. Line diagram of aggregated Core Temperature for each value of Heat Index. . 59 26. Line diagram of aggregated Core Temperature for each value of Heat Index by Activities. ................................ ................................ ................. 59 27. Scatter plot showing the association between Core Temperature and Activity Level. ................................ ................................ ...................... 60 28. Violin plot showing the distribution of Core Temperature across the four activities. ................................ ................................ ................. 62

PAGE 12

xii 29. Core Temperature versus Activity by volunteers, width of each line represents the Activity Level. ................................ ................................ ....... 63 30. Violin plot for Physiological Load of all volunteers. ................................ 64 31. Violin plot showing the distribution of Physiological Load for Female and Male volunteers. ................................ ................................ ................ 66 32. Line diagram of aggregated Physiological Load in log scale for each value of Heat Index. ................................ ................................ ............... 67 33. Line diagram of aggregated Physiological Load for each value of Heat Index. . 67 34. Scatter plot showing the association between Physiological Load and Activity Level. ................................ ................................ ...................... 68 35. Scatter plot showing the association between Physiological Load and Activity Level. ................................ ................................ ...................... 70 36. Physiological Load versus Activity by volunteers, width of each line represents the Activity Level. ................................ ................................ ....... 71 37. Violin plot for Mechanical Load of all volunteers. ................................ .. 72 38. Violin plot showing the distribution of Mechanical Load for Female and Male volunteers. ................................ ................................ ................ 74 39. Line diagram of aggregated Mechanical Load in log scale for each value of Heat Index. ................................ ................................ ...................... 75 40. Line diagram of aggregated Mechanical Load for each value of Heat Index. ... 75 41. Scatter plot showing the association between Mechanical Load and Activity Level. ................................ ................................ ...................... 76

PAGE 13

xiii 42. Violin plot showing the distribution of Mechanical Load across the four activities. ................................ ................................ ................. 78 43. Mechanical Load versus Activity by volunteers, width of each line represents the Activity Level. ................................ ................................ ....... 79 44. Violin plot for Activity Level in log scale of all volunteers. ....................... 80 45. Violin plot showing the distribution o f overall Activity Level in log scale for Female and Male volunteers. ................................ .......................... 82 46. Line diagram of aggregated Activity Level in log scale for each value of Heat Index. ................................ ................................ ...................... 83 47. Line diagram of aggregated Activity Level in log scale for each value of Heat Index by Activitie s. ................................ ................................ ...... 84 48. Violin plot showing the distribution of Activity Level in log scale across the four activities. ................................ ................................ ................. 85 49. Violin plot showing the distribution of Heart Rate for volunteer 1,2,3,4 across the four activities. ................................ ................................ ...... 86 50. Violin plot showing the distribution of Breathing Rate for volunteer 1,2,3,4 across the four activities. ................................ .............................. 86 51. Violin plot showing the distribution of Core Temperature for volunteer 1,2,3,4 across the four activities. ................................ .............................. 86 52. Violin plot showing the distribution of Posture for volunteer 1,2,3,4 across the four activities (taking absolute value for Posture). ................................ 87 53. Violin plot showing the distribution of Posture for volunteer 1,2,3,4 across the four activities. ................................ ................................ ........... 87

PAGE 14

xiv 54. Violin plot showing the distribution of Activity Level for volunteer 1,2,3,4 across the four activities. ................................ .............................. 87 55. Violin plot showing the distribution of Mechanical Load for volunteer 1,2,3,4 across the four activities. ................................ .............................. 88 56. Violin plot showing the distribution of Physiological Load for volunteer 1,2,3,4 across the four activities. ................................ .............................. 88

PAGE 15

1 CHAPTER 1: INTRODUCTION Construction is a labor intensive industry where construction workers are subjected to intensive physical stress and strain, awkward work postures, heavy weight lifting, and a range of work postures (Hartmann and Fleischer 2005) . Injuries from construction activity can be immediate or cumulative over time. This can lead to decr ease s in work productivity, inattentiveness or inability to m ake wise decision s . Demanding workload can take a toll o n both mental and physical health of a w orker. Furthermore, c ontinuous depression and dissatisfaction from work site can result in unwanted accidents and more injuries (Abdelhamid and Everett 2002) . Construction workers are exposed to ergonomic hazards, which includes dynamic movement, various awkward postures, pulling and lifting loads (Hartmann and Fleischer 2005; Schneider and Susi 1994) . They are at risk to develop musculoskeletal disorders (Forde et al. 2005; Goldsheyder et al. 2002; Griffin et al. 2000; Holmström and . However, p r evious studies have primarily been limited to one or more physical indicators of the construction worker . For example, research by (Wang and Fu 2016) and (Kirk and Sullman 2001 ) studied heart rate and demonstrated that heart rate can be used to detect physiological strain in different situations . The physical characteristics such as age, h eight and weight also impact the stress and strain the individual experiences .

PAGE 16

2 Literature Review: In this section, the author will discuss previous research focused on occupational health hazard s and the safety issues of construction workers at jobsite , as well as explore which factors have been shown to influence health and safety on a construction site. Many means for detecting these factors exist. This literature review specifically consider s the collection of live data from construction workers and its analysis to find association between the physiological me trics and environmental factors . M any construction activities require extensive physical labor over long period of time in adverse weather conditions. Construction activities often include heav y lifting, pulling and pushing and various awkward posture s. (Hartmann and Fleischer 2005) suggests that construction work, safety and productivity are connected with each other and can influence individual construction worker health . Measuring p hysical strain can show how the s occupational health is affected , and , consequently , help to ma intain and promote safety on the job site (Abdelhamid and Everett 2002; Bouchard and Trudeau 2008; Garet et al. 2005) . Correlating the he alth of a construction worker with physiological measures to detect physical strain is a relatively new concept. One of the first studies in this area was done by (Kirk and Sullman 2001) . They used heart rate as the main indicator to determine physical strain of cable hauler choker setters in New Zealand. Results ident ified a correlation of heart rate to physical strain and overall assessment of workload for a specific task. This study questioned the ability to collect accurate real time data when the workers are in motion without causing personal discomfort. A

PAGE 17

3 similar study by (Abdelhamid and Everett 2002) added oxygen consumption to heart rate data to determine the performance of 100 construction workers doing moderate to heavy work. The data showed that the average oxygen uptake was 0.82 L/ min ( ± 0.22 L/min) and the average heart beat was 108 beats / min ( ± 17 beat) min) . These measurement s were compared to standard guidelines for acceptable levels of physical performance for specific industrial environment . The results indicated that 20 40% of the regularly exceed threshold levels determined by standard published guidelines for ma nual work . E xceeding the threshold limits cause workers to be more prone towards inattentiveness , decreased productivity, poor judgment, accidents and injuries in job site. (Roja et al. 2006) conducted a study on workers from the heavy civil industry which included ten road construction and maintenance workers and ten pavers who belonged from the age group 20 60 years. T h e study measured the physical demands of road construction work and estimate d muscle fatigue. Average metabolic energy consumption fo r road construction and repairing works were recorded as 8.1 ± 1.5 kcal/min , and 7.2 ± 1.1 kcal/min for paving. Their findings, based on average metabolic energy consumption , indicated that road construction work requires extreme manual labor, compulsive working posture, and continuous arm and leg movements. In the same study t hey monitored the heart rate, posture and muscle tone for a week and suggested that work related musculoskeletal problems could be common f or these workers . (Taylor et al. 2010) claimed that the symptom s of musculoskeletal disorders (MSD) are common for young construction workers . The chance of MSD increase with the increasing years and women are more effected than men by MSD. Early prevention

PAGE 18

4 strategies can reduce disability associated with MSD . Another similar study was conducted on highway tunnel construction workers by (Tak et al. 2011) focusing hazards. They found that non neutral trunk postures and, in some operations, kneeling and squatting are fairly common. These postures pose risk for musculoskeletal disorders, effecting the back and knees in particular. Construction w orkers in some part of the world are exposed to extreme heat . Climatic Heat stress can be fatal. (Rowlinson et al. 2014) studied factors affecting climatic heat stress and identified thr ee ways of reducing heat stress in construction sites : 1) Control of climatic heat stress exposure through use of an action triggering threshold system, 2) control of Continuous Work Time with mandatory work rest routine, and 3) allow workers to follow sel f pace regimes. However, each construction site is unique and each job site should have its own set of heat stress guidelines. (Bates and Schneider 2008) conducted a study on construction workers in thermally extreme conditions in United Arab Emirates ( UAE ) and monitored temperature, fluid intake, urine specific gravity and heart rate. The y concluded that people working in extreme conditions can avoid fatigue when provided with plenty of fluid and are allowed to self pace work . Other research studied th e impact of the climate in Hong Kong on construction workers . In Hong Kong, t he industry demands long and irregular hour s . (Yi and Chan 2015) established a Monte Carlo simulation based method to find the optimal w o r k pattern. To sustain the climatic adversity the authors propose d a schedule with a direct work rate of 87.8% of the entire work time as an ideal work pattern for the construction workers in Hong Kong. Job rotation was also

PAGE 19

5 considered to be an effective strategy to promote health in these working conditions. Finally, c onsistent, period breaks are recommended to minimize occupational health hazard s and maximize productiv ity . Another strategy to mitigate climatic heat stress as suggested by (Yi et al. 2016) is to use an early warning system that prompts signals to worker in distress. This system is developed with the help of smart bracelet and a smart phone that collects data from the construction worker. Artificial neural Network is used to obtain the final signals. The heat strain here is calculated from a subjective index rating. The proposed system can be used to protect the well being of the as the new algorithm helps to send out early warning alerts to the construction workers . A strong connection exists be tween physical performance and hydration. Dehydration leads to changes in cardiovascular, thermoregulatory, metabolic and central nervous system function that hinders the performance , and h eat exposure along with dehydration can be fatal situation for cons truction workers on the job. (Murray and Murray 2017) studied the relationship between hydration and physical performa nce. Practical guidelines at construction site encourages consumption of water without overdrinking. Reducing sweating rate, taking frequent rest break, drinking sufficient volumes of fluid are the simplest way s to improve physical performance. Recently, significant research has focused on the roll technology can play in monitoring worker health and safety. (Cheng and Teizer 2013) s tudied the opportunity to use video camer as on construction job sites to track worker dynamics.

PAGE 20

6 (Marx and König 2011) introduced physical monitoring system (PMS) to estimate construction physical strain based on heart rate and acceleration and found out two of the three PMS system are reliable based on different activities and testing. The PMS devices selected f or this study are Zephyr BioHarness (BH) BT, Zephyr HxM, and Hidalgo EQ 01 . (Hwang and Lee 2017) have shown the association between Heart Rate Reserve percentages (%HRR) and physiological demand. This study revealed physical demand is dependent largely on changing working conditions, demand of the task, age, working conditions and working patterns. (Romagnol 2010) showed how real time data analysis can very effe ctive and efficient during emergency and extreme condition of construction workers. (Buller et al. 2010) discuss ed the importance of estimation of core temperature of human body during construction activity. Another research used real time location sensing (RTLS) and physiological status monitoring (PSM) to monitor the posture and physical status of the workers. (Cheng et al. 2013) (Hwang and Lee 2017) verified the importance of continuous physical measurement to detect any significant and sudden change in their physical demand. Smart mo bile phones, medical sensors, wireless communication system, portable trackers and health monitoring systems have been used to track continuous real time data for medical patients , workers and athletes . (Banos et al. 2014) discussed a PhysioDroid that monitor s data remotely based on vital signs like electrocardiogram, heart rate, respiration rate, skin temperature, and body motions. The versatility of the device is a key factor. Monitoring construction physiological conditions , however, can differ from sports, science and medicine. (Gatti et al. 2014) investigated the validation of PMS s for construction based on two physiological parameters such as

PAGE 21

7 heart rate and breathing rate. Results indicated that PMS can successfully analyze association between physical strain and task level, productivity assessment and safe unsafe behavior of construction workers . (Scheer et al. 2014) collected heart rat e data, weather data and video recordings to establish the relation that PSM technology certainly indicates overexertion, potential fatigue, illness , and cardiovascular overload. Another study involving PSM and heart rate data tried to measure the physiol ogical cost of concrete construction activities (Lee and Migliaccio 2016) . H eart rate was observed to be a reliable predictor of physiological stress level. Physiological cost of construction activities were calculated from HR indexes and descriptive statistics , including mean, standard deviations , Physiological cost comparison revealed concrete construction activities to be moderate and comparable to steelwork. Placing and vibrating concrete produced more physical strain , and requires more onsite personal safety mea sures. However, the study showed that p hysiological cost varies over time. The higher cost are observed during the mid day and at the beginning and end of each week. Another study on roof top construction was conducted to estimate Total Worker H ealth (TWH) by (Lee et al. 2017) . TWH merge occupational health hazards with The study focused roofing workers since they frequently encounter high on duty health and safety risks and poor off duty lifestyles. Th e study also validate d the reliability and usability of wearable sensors. (Wang et al. 2017) proposed the use of a wireless and wearable electroencephalography (EEG) system t o process the brain signals . Results from their on si te experiment show EEG signals such as frequency, power spectrum

PAGE 22

8 density and spatial distribution can effectively detect the construction perceived level of risk. (Zhou and Ding 2017) proposed an Internet of Things (IoT ) based safety barrier alert system to avoid hazardous injury for underground construction . This technology has been implemented in the construction of the Yangtze River crossing metro station to test its performance. Th e system generate s timely early warning alerts under critical environment using spatial temporal barrier . (Aryal et al. 2017) i ncorporated a heart rate monitor, infrared temperature sensors device and an ECG sensor enabled live data monitoring for 12 individuals. The ir results suggest that data combined from all the sensor s were 8 2% accurate in predict ing p hysical fatigue. (Awolusi et al. 2018; Yu et al. 2017) also validated the use of real time data based for physiological monitoring, environmental sensing, proximity detection and location tracking analysis on construction workers W orker injuries and accidents, equipment damage directly contributes to disrupted work schedules, worker compensation, insurance hike, financial loss. Advanced safety efforts in construction site can avoid or minimize all these costs. (Abudayyeh et al. 2006) concluded how the management committee and safety managers working as a team can improve the overall conditions by focusing on engineering improvements along with emphasizing human behavior commitment from the entire organization. Worker training programs can i mprov e safety performance and enhanc e work efficiency. (Teizer et al. 2013) in their study proposed a new approach to integrate

PAGE 23

9 real time data from construction job site and incorporate them in prior education and training programs. With real time location tracing and visualization technology , they helped workers to identify safety issue that they were not aware of due to their natural working e nvironment and personal habits . Combination of real time data tracking and visualization technology was shown to be advantageous in terms of improving occupational health hazards. Construction workers are frequently required to work at high heights and managing fall from heights (FFH) is one of the essential part of construction safety management. Numbers show that FFH is one of the leading cause for serious injuries and fatalities across the world. (Guo and Goh 2017) identified the causes of FFH and develop a universal ontology model that would benefit all construction workers from building and construction industries . (Fang and Dzeng 2017) designed a fall detecting method based on a hierarchical threshold based algorithm to identify and avoid fall accidents. (Yang et al. 2017) proposed a sensing device approach that understands the gait irregularities to identify physical fall in a construction site. In this study , the wearable sensory device s used to collect kinematic gait data were attached to the ankle . The cumulative gait data was analyzed to obtain gait abnormality score r epresenting the hazardous situations in job site. (Choi et al. 2017) investigate d how to motive construction workers to put on wearable devices , in this case, smart vest with GPS device on them and a wristb and type wearable activity tracker , and found that workers were more likely to wear such devices if they were aware of the potential benefits. General Behavior Based Safety (BBS) is based on visual recording and manual observation, which is an impractical

PAGE 24

10 solution for the construction jobsite. However, (Yu et al. 2017) suggested a theoretical framework using an image skeleton based parameterized method. This research improves the effectiveness of the construction safety management. (Choe and Leite 2017) formulated a 4 dimensional (4D) construction safety method which is specific to space, time , and location. V isual safety materials were used to enhance safety communication at jobsite. A particular case study revealed that risky activity, days and zones can be distinguished from saf er activity, days and zones. This research helps the safety personnel with a proactive, job site specific safety plann ing model that helps to manage the construction site. Finally, (Hwang and Lee 2017) successfully converted the heart rate to heart rate reserve (%HHR) to set a threshol d on acceptable amount of time a construction activity needs a specific amount of % HRR. Once the % HRR is proved to be consistent, it can be used in different construction equipment to measure heart rate as % HRR. Hwang and Lee (2017) established successf ul relationships between % HRR and influencing factors to show significant high physical demand during construction work. Despite significant research into construction worker health and safety, and re cent focus on using technology to assess spatial and physical conditions on construction sites, a critical need remains for monitoring and collecting physiological data of construction workers. This research addresses this need by collecting physiological and environmental data o n US Air Force Academy cadets performing four different construction activities during a summer field course in Colorado Springs,

PAGE 25

11 Colorado . The author using the data set to explore and statistically compare the impact of construction activity on individual worker s under similar environmental conditions across activities .

PAGE 26

12 CHAPTER 2: METHODOLOGY AND RESEARCH E quipment: V arious sensors have been used and validated in the past to monitor physical activities . Physiological status monitors (PSMs) is a readily available and effective device for monitor ing heart rate, breathing rate and acceleration (Zhen Wang & Shan Fu, 2016). However, heart rate is significantly related to breathing rate and body acceleration , a nd is also effected routine, food habits and other personal factors. While heart rate can detect fatigue , with more indicators there will be better chance to assess performance ability. Research has shown tha t e nvironmental condition like temperature and humidity also impact fatigue . For this study , the author used Zypher Bioharness system to monitor volunteers . Zephyr Technologies specializes in remote physiological monitoring and location tracer without hindering the flexibility and freedom of the any individual. This technology has been used to collect data from professional sports person, special force, rescue workers , medical field and in other research studies . The Zephyr Bio harness system are most commonly used in sports and sport medicine to optimize the work rest ratio, allow ing for a better heart rate recovery regime and optimized physical performance . ( Zephyr 2012, 2013, 2016)

PAGE 27

13 Different Components of the Bio harness Device: The following Zephyr Bio harness components were used for this study . 1. Bio harness Strap an approximately one inch wide strap worn across the chest with the puck inserted in the desired slot. There is a compr ession strap which allows the puck to be on exact position against the sternum. The correct position of the puck is along the edge of the strap under the arm pit, or slightly to the rear. 2. Puck the main sensory device that is pressed by the compression str ap against the sternum to collect data. 3. Puck docking station holding device that charges the pucks and download s the data. 4. OmniSense Software a proprietary software which allows viewing of live data and recorded data. It also includes capabilities to ana lyze the data using predefined algorithms. M etrics : This research applied the following metrics : heart rate, breathing rate, core temperature, mechanical load, physiological load and posture. The following are the description are based on and extend those provided by the manufacturer (Zephyr 2013) .

PAGE 28

14 Table 1 : Pilot data metrics Metric Description Units Notes Heart Rate It is measured as the number of heart beat per minute. Heart beats per minute The measure of Heart rate is analyzed from the 250Hz Echocardiogram (ECG) data Breathing Rate It is measured as the number of breathes per minute. Breathing per minute The sensors inside the zephyr puck detects breathing by the expansion and contraction of our torso. Posture It is measured as the change of angle of any individual in comparison with the gravity. Degree from vertical position. When any individual is standing straight the measurement is zero. Forward and backward leaning accounts for positive and negative values. Core Temperature The core temperature is calculated based on the heart rate , skin temperature and a predefine formula by Zephyr system Degree Centigrade T he accuracy of this estimate and have also demonstrated that such a computational measurement can indicate physical stress before an individual reaches an unhealthy state (Buller and Hoyt 2008) . Mechanical Load Mechanical intensity over time, where mechanical intensity is a measure of instantaneous effort based on accelerati on Unitless M easured based on the movement of an individual and acceleration. Physiological Load Physiological intensity over time, where physiological intensity is a measure of instantaneous effort based on heart rate Unitless M easure d based on cardiovascular output. Activity Level The activity metric is a collective measurement of the accelerometer given in velocity magnitude units (VMU) ranging between 0g to 16g. Acceleratio n due to gravity The averages of the three axial acceleration magnitudes over the previous 1 second epoch, sampled at 100Hz.

PAGE 29

15 Data Collection: This study was conducted on 10 cadets from the 2017 summer program at the Field Engineering and Readiness Laboratory (FERL) at the United States Air Force Academy . The cadets self volunteered for this study. International Review Board protocols for research involving human subjects at both University of Colorado Denver and US Air Force Academy (USAFA) were completed and thoroughly complied to. FERL has a 3 week long on experiences in surveying, construction methods, and ("Civil Engineering Practices Field Engineering Cadet Handbook," 2017) . The program is designed such that all cadets are exposed to classroom and field activities that give them an experience about civil construction activities. The cadets were performing all the activities at a high elevation. We did not consider any additional effect of high altitude on the physiological metrics of the cadets. The cadets are grouped in different individuals . The study was conducted over three weeks where the volunteer subjects were exposed to various construction work in re gard to civil engineering like concrete placemen t, surveying etc . It is important to note that all the volunteers were strictly following similar food and sleep regiments while staying at the academy. The schedules for each flight was different from one another. However the schedule is designed such that all flights had to perform all activities over the course of three week. For our study we mainly focused on 4 activities: concrete placement,

PAGE 30

16 heavy equipment, surveying and asphalt paving. For simplification of our analy sis we numbered the different activities which are included in Table 2 . The data collected from the ten volunteers comprised of 8 male volunteers and 2 female volunteers. All the cadets were within the age range of 20 22. Prior to the study w e collected the individual characteristics of the cadets which included weight, height , and age. This data was imported to analyze the physiological status for each volunteer (Zephyr 2016) to the software also requires input of a fitness level for each individual. Physiological status monitoring is a way to collect and record the vital signs of the volunteers. It is a nonintrusive method which can occur in real time. Since all of the volunteers were cadets in the United States Ai r force and they were in their early twenties, each was assigned an eight out of ten for fitness. Data were collected over three weeks. However there were some missing data when volunteer s forgot to switch on t he device. Moreover there were some inconsist encies when the straps were loose and di d not fit the cadets properly. To maintain the quality of the data , they w ere filtered and treated before running the analysis. Omnisense software generates a metric called Heart Rate Confidence (HRC). HRC (%) is calculated based on the electrocardiogram (ECG), ECG noise and worn detection. The threshold for accepting data was based on HRC of 80% or more, based representative.

PAGE 31

17 Assumptions: 1. For our research we assumed that all our cadet volunteers (aged between 20 22 years) have a high level of fitness. The United States Air force Academy at Colorado Springs is situated at an elevation of 7258 feet. All the activities were performed in this high elevation. 2. They volunteers were from different flights performing similar activities over the three week summer program. We assumed all the volunteers were doing exact similar activities on different days. 3. We assumed that the Zephyr Bioharness Pucks were pre calibrated. Table 2 : Construction Activities Activity Activity Number Description Concrete Placement (CP) 1 Prepare a site, set frame work and reinforcement, place concrete, and take samples and perform slump test Heavy Equipment (HE) 2 Operate construction equipment including an excavator, scraper, bulldozer, loader, and paving machine. Surveying (S) 3 Use total station to measure distance and horizontal/vertical angles. Lay out a calculated location for concrete slab pour and use measuring techniques to plot land data. Asphalt Paving (AP) 4 Place a section of road using approximately 20 tons of asphalt.

PAGE 32

18 Table 3 : Characteristics of Individual Participants Volunteer Gender Height (feet, inches) Weight (pounds) Age (years) 1 M 5'11" 154 22 2 M 6'2" 187 21 3 M 6'0" 183 21 4 M 6'0" 170 20 5 F 5'11" 175 21 6 F 5'2" 145 20 7 M 5'11" 171 20 8 M 5'7" 170 21 9 M 6'4" 180 20 10 M 6'4" 180 22 Methods: Data filtration and organization: The B ioharness puck sy s tem has sufficient memory to store 36 hours of data at one minute interval. The data can be viewed either in real time using a blue tooth connection to OmniSense Live software, or can be viewed after downloading the OmniSense Analysis software. For our study , the data for all the ten volunteers at one second interval we re downloaded each day manually . Statistical software R was used t o run all the statistical analysis. Omnisense software allows user to download excel files for further study and research purpose which is beyond the scope of Omnisense software. These excel files can be imported using R and subsequently analyzed. O ur research focused particularly on four activities 1. Concrete 2.Heavy equipment 3. Surveying 4. Asphalt . The first step is to filter the day we need the dat a

PAGE 33

19 for our analysis, since there were o ther activities that the volunteers performed over the three week s of summer program. Once the days were selected , all the data w ere data frame was used to run the analysis for our research. The analysis part comprise s of two major section s . The first part includes all the filtration and organization while the other part is mostly about fitting models to answer our research questions. The steps for the organization part are as follows. Step 1 : The data frame was filtered by HRC %. Any row in the data matrix with HRC below 80 was removed. Step 2: Only the column s corresponding to the metrics that we are interested in (e.g., heart rate, breathing rate etc. ) were kept and the rest were removed. S tep 3 : The data set contains observations at one second interval s . For the sake of simplification and convenience we co llapsed the data for every second to data for every minute by taking a summary measure within each minute. Depending on the type of metric differen t summary measures were decided for different metric s . For most of our metric s, average was used to collapse data within each minute. However, mechanical and physiological loads are cumulative load that builds up over time. Thus the gain for each minute (maximum minimum) was calculated for these two metrics . Table 4 s hows the different summary measures. Finally, before fitting any model or further analysis the data set was checked for any abnormal values or outliers.

PAGE 34

20 Table 4 : List of summary measures to collapse data for different metrics Metric Summary measure to collapse data within every minute Heart Rate Average for each minute. Breathing Rate Average for each minute. Posture Average f or each minute. Core Temperature Average for each minute. Mechanical Load Gain for each minute. Physiological Load Gain for each minute. Activity Level Average for each minute. Weather Data : Weather parameters influence the physiological metrics for a n y construction worker . Previous studies have indicated that exposure to the extreme temperature can influence the physical strain of a construction worker working outdoors . We collected temperature and humidity data for the time period when the volunteers were performing their activiti es. The source for our data was United States Air Force Academy weather station as reported on the Weather Underground website ( https://www.wunderground .com/history ). After collecting the temperature and relative humidity data , the researchers evaluated the Heat Index (HI) based on Ersatz version of the heat index equation (Rothfusz 1979) g iven by Equation 1 W here , T=am bient dry bulb temperature (°F)

PAGE 35

21 R= Relative Humidity (integer Percentage) The activity schedule was divide d into morning and afternoon shifts and two value s of Heat Index were available for each day. Table 5 sho ws the days a nd the heat index values for morning and evening session.

PAGE 36

22 Table 5 : Timetable of Weather Conditions and Construction Activity by Volunteer DAYS TEMPERATU RE ( ° F) RELATIVE HUMIDITY (%) HEAT INDEX CONCRETE HEAVY EQUIPMENT SURVEYING ASPHALT 6/5/2017 DAY 1 65.3 52.83 78.32 V 6,10 F B 74.4 28.2 76.61 6/6/2017 DAY 2 64.6 49.5 78.99 V 8 F C 64.7 56 78.30 6/7/2017 DAY 3 64.3 58.83 78.09 V 9,5 F D 70.1 47.8 77.02 6/9/2017 DAY 5 75.4 23.17 76.48 V 1,2,3,4 F A 85.1 12.2 81.93 6/12/2017 DAY 6 69.9 54 76.64 V 9,5 F D 83.3 13.8 80.58 6/13/2017 DAY 7 65.9 19.5 73.22 V 1,2,3,4 F A 76.6 6.4 75.28 6/14/2017 DAY 8 68 20 73.75 V 1,2,3,4 F A V 6,10 F B 79 13.4 77.59 6/15/2017 DAY 9 71.2 23.83 75.28 V 6,10 F B V 8 F C 83.8 11.4 80.94 6/16/2017 DAY 10 72 28.5 76.09 V 8 F C V 9,5 F D 87.1 10.8 83.52 6/196/201 7 DAY 11 67.9 38.33 77.26 V 6,10 F B 80.8 25.2 79.36 6/20/2017 DAY 12 73.2 37.83 77.00 V 8 F C 88.3 15 84.52 6/21/2017 DAY 13 80.3 29.83 79.35 V 9,5 F D 87.1 17 83.58 6/23/2017 DAY 15 50.6 82.17 88.13 V 1,2,3,4 F A

PAGE 37

23 Model fitting: Multiple L inear regression model is predominantly used to model relationship among two or more explanatory variables and a response variable by fitting a linear equation to the observed data . The equation for a multiple linear regression with p explanatory variables has the form , ( Equation 2 ) w here , = response /dependent variable for th sample Estimated regression coefficients , = value of the explanatory/ in dependent variables for th sample , = error (residual) In case of a multiple linear regression it is assumed that all are independent. However, for our data there are repeated observations over time for an individuals which cannot be assumed to be independent. An appropriate way to model such repeated measures data across time is the use of linear mixed effects model. The Linear Mixed Effect Model (LMM) comprises of both fixed effects and random effect s . The random effects take care of the correlation between the repeated observations. For our study we aimed to figure out the effect of the independent variables such as gender, heat index, activ ity level and activity on the me tr ic s. The LMM is represented as follows.

PAGE 38

24 , ( Equation 3 ) w here , = Value of the response for the th replication for th individual = individual specific random effect for th individual = explanatory/independent variables, = error (residual) The model assumes , and they are independent. Specifically for our data, t he model can be w rit ten as ( Equation 4 ) = Value of the response for the th replication for th individual Activity2= Activity3= Activity4=

PAGE 39

25 Gender= H.I= Heat Index A.L= Activity L evel = individual specific random effect for th individual = error (residual) Figure 1 : Correlation Heat Map between different metrics . Prior to deciding about our model, w e computed the correlations among the different me trics to understand their inter relations as well as to decide which variables to include in the model. Figure 1 shows the correlation between all the metrics for our study. This heat correlation coefficient. The color bar in the figure shows the scale for strong and weak correlations.

PAGE 40

26 Although Activity L evel is one of our metrics of interest, we also include it in the model s (except when the response is Activity L evel) as an explanatory variable since it has high corr elation with other variables and can be thought of as a causal variable that can influence the responses. It can influence other metrics but not the other way around. Since Activity Level has a skewed distribution, we use log transformed values when consid ering it as the response variable in our models. The LMM is used to answer most of our research questions. The research questions and the corresponding null hypotheses are as follows. Q1. For each metric, is the outcome statistically similar for all activities? vs : is not true. Q2. For each metric, is there any statistical difference in the outcome between men and women ? vs Q3. For each metric, is the outcome statistically similar across different heat Index? vs Q4. For each metric, is the outcome statistically similar across different activity levels ? vs The hypotheses can be tested using statistical test of significance. T test s are used to test the hypotheses for Q2, Q3 and Q4, while a Likelihood Ratio Test (LRT) is used to test the hypothesis for Q1. If a significant difference between different

PAGE 41

27 activities is found for Q1, we then followed up using pairwise comparisons ( t test ) to test which activities have significantly different outcomes . Since multiple hypotheses are tested, we need to adjust the raw p values using methods for multiple testing adjustment. We used Bonferroni correction (Dunn 1961) for this purpose. The metrics Physiological Load and M echanical Load are observed to be not only positively skewed, but also zero inflated. Therefore, regular LMM is not appropriate for these metrics. Instead , w e use Co m pound Poisson Generalized Mixed Model (CPGLMM), a model specially designed for such data (Zhang 2014) . Description of results The following is a list of the visualization techniques and a summary of the statis tic al data presented for each metric in the Results Section to follow . Results are presented using violin plots, as well as tables listing T test p values and pairwise comparison using Bonferroni a djusted p value s . For each metric the following information is provided: 1. Violi n Plot Violin plots visualize the distribution of the data . T hey are vertically symmetric presentations of the probability distribution of the response variable as estimated fr om the data. The mean and the mean ± 1 standard deviation range are also shown. The following sections present the results for each metric in succession. 2. Table showing s tatistical analysis of effect of Gender, Heat Index, Activity Level and Activity on Me trics . The following generic table explains in details the representative symbols along with their meaning and description. To test

PAGE 42

28 for statistical differences a T test was performed for the metric data across all volunteers. Table shows the fitted LMM and the p values corresponding to the hypotheses to be tested for Q1, Q2, Q3 and Q4 . Column 1 rep resents the effects and column 4 represents the estimates of the corresponding regression coefficients ( ) . The test statistics and the Bonferroni adjusted p values are reported in column 5 and 6 respectively. Table: Description of statistical data analyzed and presented for each metric. Effect Coefficient Interpretation Estimat e Test statistic P value Gender Average difference in HR between males and females. Heat Index Average increase in the HR when HI increases by one unit. Activity level Average increase in the HR when AL increases by one unit. Activity Average difference in HR for activities HE & C Average difference in HR for activities S & C -Average difference in HR for activities A & C Note: Fit of LMM for Heat Rate showing the effect of Gender, Heat Index, Activity Level and Activity on Heart Rate. The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated using likelih ood ratio test. Note that in interpretation column the Y and assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the difference/increase is 0. 3. Line diagrams A line diagram is a graph that connects a series of points by drawing line segments between them. These points are ordered in one of

PAGE 43

29 their coordinate (either the x coordinate or y coordinate ) value. Line diagrams are commonly used to represent trends in data. 4. Scatter plots A graph in which the values of two or more variables are plotted along two axes, the pattern of the resulting points reveal s if there is any correlation present in the data set. 5. Pairwise Statistical C omparison table This table consist of pairwise follow up comparison between different activities. Each cell in the table shows Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwise comparisons of p values indicates statisti cally distin ction in metric profiles among different activities. Table: Comparison between different activities for Metric Activity Concrete Heavy Equipment Surveying Asphalt Concrete Heavy Equipment Surveying Asphalt Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison.

PAGE 44

30 CHAPTER 3: RESULTS Results The following results address the research question s. The following sections present the results for each metric in succession. Heart Rate: Table 6 shows the Heart rate distribution s for the 10 volunteers. Different profiles are evident for different Volunteer s . As shown, s ome Volunteer have higher mean Heart rate s than others. To test f or statistical differences a t test was performed for the Heart Rate data across all volunteers as shown in Table 7 .

PAGE 45

31 Figure 2 : Violin plot for Heart Rate of all volunteers. Table 6 : E ffect of Gender, Heat Index, Activity Level and Activity on Heart Rate Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 1.074 t=0.177 1 Heat Index Average increase in the HR when HI increases by one unit. 0.096 t= 3.491 0.033 Activity level Average increase in the HR when AL increases by one unit. 103.081 t=81.397 0 Activity Average difference in HR for activities HE & C 1.472 =309. 31 Average difference in HR for activities S & C 3.841 Average difference in HR for activities A & C 0.594 Note s : Fit of LMM for Heat Rate showing the effect of Gender, Heat Index, Activity Level and Activity on Heart Rate. The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated using likelihood ratio test. Y and assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the difference/increase is 0.

PAGE 46

32 The p values indicate that Heat Index, Activity Level and Activity are significantly associated with Heart Rate (p 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( ) and the residual variance ( ) are estimated to be 89.30 and 104.86. Figure 3 sh ow s the frequency distribution of Heart Rate for females and males . It visually confirms the results of Table 6 that Heart Rate differences are not statistically significant across gender. Figure 3 : Violin plot showing the distribution of overall Heart Rate for Female and M ale volunteers.

PAGE 47

33 Figure 4 and Figure 5 depict the relationship of Heart Rate to Heat Index with each colored line representing an individual volunteer. Although the p values indicate significant associations significant, it is difficult to draw clear conclusions about patterns in the relationship between Heart Rate and Heat Index. For example, implies that with a change of Heat Index from its minimum observed value 73.22 to its maximum observed va lue 88.36, the Heart Rate decreases only by 1.46, on average. Since the number of observations is very large, the p values can be very small even when there is a small, practically negligible effect size. W hile a general trend may be that Heart Rate declin es as Heat Index rises (as most clearly demonstrated by V1 V4) , more research is recommended to further analyze this relationship. Figure 4 : Line diagram of aggregated Heart Rate for each value of Heat Index .

PAGE 48

34 Figure 5 : Line diagram of aggregated Heart Rate for each value of Heat Index by Activities. Figure 6 p lots Heart rate vs. Activity level, colored by Activity and suggests a positive correlation (upward trend) exists between Heart Rate and Activity L evel.

PAGE 49

35 Figure 6 : Scatter plot showing the association between Heart Rate and Activity Level. Figure 6 p lots Heart rate vs. Activity level, colored by Activity and suggests a positive correlation exists between Heart Rate and Activity L evel. Next , pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Heart Rates. Results are summarized in Table 7 .

PAGE 50

36 Table 7 : Comparison between different activities for Heart Rate Activity Concrete Heavy Equipment Surveying Asphalt Concrete 4.07E 08 0 1 Heavy Equipment 4.07E 08 0 0.301 Surveying 0 0 0 Asphalt 1 0.301 0 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwise comparisons of p values indicate that Concrete Heavy Equipment, Concrete Surveying, Heavy Equipment Surveying, Heavy Equipment Asphalt result in statistically distinction Heart Rate profiles from one another . Finally, the Heart Rate during surveying is the most distinct compared to the other three Activities. Figure 7 shows that Heart Rate is, on average, lowest during surveying and spikes highest during Concrete Placement.

PAGE 51

37 Figure 7 : Violin plot showing the distribution of Heart Rate across the four activities. To understand the relationship of Heart Rate with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Heart rate for each volunteer with the width of the lines showing the amount of Activity Level in Figure 8 . As expected, the lines become wider towards the top of the figure, but that is not always the case. For exam ple, for volunteer 4, we clearly see the pattern Concrete > Heavy Equipment > Surveying > Asphalt, despite higher activity levels with Asphalt compared to Surveying. Other similar patterns are

PAGE 52

38 observed, but there is no consistent pattern across all volunte ers. This implies that there is interaction between volunteer and Activity and including that in the model might be ideal. However, since Activities are modeled by fixed effects and volunteers are modeled by random effects, it can only be achieved using a highly complicated model which is beyond the scope of this thesis . However, t here is no consistent pattern across all volunteers , which implies more research is recommended to analyze these relationships. Figure 8 : Heart rate versus Activity by volunteers , width of each line represents the Activity Level. Breathing Rate: Figure 9 shows the Breathing Rate distributions for the 10 volunteers. Different Volunteers have different Breathing Rate profiles . As shown, s ome Volunteer have higher mean Breathing R ate s than othe rs.

PAGE 53

39 Figure 9 : Violin plot for Breathing Rate of all volunteers. To test for statistical differences a t test was performed for the Breathing Rate data across all volunt eers. Table 7 sho ws the results of the statistical testing. The p values indicate that Heat Index, Activity Level and Activity are significantly associated with Breathing Rate (p 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( ) and the residual variance ( ) are estimated to be 1.86 and 8.86. Figure 10 Figure 13 show the associations between Breathing Rate and the independent variables. Figure 10 shows the distribution of Breathing Rate by Gender. It visually confirms the results of Table 8 that Bre athing Rate differences are not statistically significant across gender.

PAGE 54

40 Table 8 : Effect of Gender, Heat Index, Activity Level and Activity on Breathing Rate Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 0.666 t= 0.753 1 Heat Index Average increase in the HR when HI increases by one unit. 0.030 t=3.773 0.011 Activity level Average increase in the HR when AL increases by one unit. 16.238 t=44.11 7 0 Activity Average difference in HR for activities HE & C 0.026 =70.5 5 Average difference in HR for activities S & C 0.483 Average difference in HR for activities A & C 1.033 Note s : Fit of LMM for Breathing Rate showing the effect of Gender, Heat Index, Activity Level and Activity on Breathing Rate. The p values for Gender. Heat Index and Activity Level are calculated using t te st and the p value for Activity was calculated using likelihood ratio test. Y and assumes that the values of all other variables remain the same. The test statistic and the p value s are related to the hypothesis that the difference/increase is 0.

PAGE 55

41 Figure 10 : Violin plot showing the distribution of overall Breathing Rate for Female and Male volunteers. Figure 11 and Figure 12 depict the relationship of Breathing Rate to Heat Index with each colored line representing an individual volunteer. Although the p values are significant, it is difficult to draw clear concl usions about patterns in the relationship between Breathing Rate and Heat Index. While a general trend may be that Breathing Rate declines as Heat Index rises, more research is recommended to further analyze this relationship .

PAGE 56

42 Figure 11 : Line diagram of aggregated Breathing Rate for each value of Heat Index. Figure 12 : Line diagram of aggregated Breathing Rate for each value of Heat Index by Activities.

PAGE 57

43 Figure 13 plots Breathing R ate vs. Activity level, colored by Activity and suggests a positive correlation (upward trend) exists between Breathing Rate and Activity Level. Figure 13 : Scatter plot showing the association between Breathing Rate and Activity Level.

PAGE 58

44 Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Breathing Rates. Results are summarized in Table 9 . Table 9 : Comparison between different activities for Breathing Rate Activity Concrete Heavy Equipment Surveying Asphalt Concrete 1 1 Heavy Equipment 1 1 Surveying 0.0002 Asphalt 1 1 0.0002 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwise comparisons of p values indicate that Concrete Surveying, Heavy Equipment Surveying, Surveying Asphalt result in statistically distinction Breathing Rate profiles from one another. Finally, the Breathing Rate during surveying is the most distinct compared to the other three Activities. Figure 14 visual ly confirms that Breathing Rate differs by Activity with Surveying having the lowest average Breathing Rate and Asphalt having the maximum range of Breathing Rate . However, it is difficult to discern which activities are statistically distinct.

PAGE 59

45 Figure 14 : Violin plot showing the distribution of Breathing Rate across the four activities. Finally, to further explore the relationship of Breathing Rate with the Activities also accounting for the difference between volunteers and the difference in Activity Level, Figure 15 is a plot of Activity versus Breathing Rate for each volunteer with the width of the lines showing the amount of Activity Level. This figure reveals how the Breathing Rate varies based on the choice of Activity for different volunteers. As expected, the lines become wi der towards the top of the figure, but that is not always the case. For example, for volunteer 10, we see the pattern Asphalt > Surveying >Heavy Equipment, despite having similar Activity Levels. However, there is no consistent

PAGE 60

46 pattern across all volunteers, which implies more research is recommended to analyze these relationships . Figure 15 : Breathing rate versus Activity by volunteers , width of each line represents the Activity Level. Posture: Figure 16 shows the Posture distribution s for the 10 volunteers. Different profiles are in Posture based are evident for different Volunteer s . As shown, s ome Volunteers have higher mean Posture than others.

PAGE 61

47 Figure 16 : Violin plot for Posture of all volunteers. To test for statistical differences a T test was performed for the Posture data across all volunteers as shown in Table 10 . The p values indicate that Heat Index, Activity Level and Activity are significantly associated wit h Posture (p 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( ) and the residual variance ( ) are estimated to be 11.2004 and 124.5344 . Figure 17 Figure 20 show the associations between Posture and the independent variables. Figure 17 shows the distribution of Posture by Gender. It visually confirms the results of Table 10 that Posture differences are not statistically significant across gender .

PAGE 62

48 Table 10 : Effect of Gender, Heat Index, Activity Level and Activity on Posture Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 3.657 0.770 1 Heat Index Average increase in the HR when HI increases by one unit. 0.043 0.806 1 Activity level Average increase in the HR when AL increases by one unit. 14.158 5.664 Activity Average difference in HR for activities HE & C 5.330 =408. 565 Average difference in HR for activities S & C 3.274 Average difference in HR for activities A & C 2.523 Note s : Fit of LMM for Posture showing the effect of Gender, Heat Index, Activity Level and Activity on Posture . The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated using likelihood ratio test. Y a nd assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the difference/increase is 0.

PAGE 63

49 Figure 17 : Violin plot showing the distribution of overall Posture for Female and Ma le volunteers.

PAGE 64

50 Figure 18 and Figure 19 depict the relationship of Posture to Heat Index with each colored line representing an individual volunteer. Although the p values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Posture and Heat Inde x. While a general trend may be that Posture declines as Heat Index rises this is not, necessarily intuitive , and more research is recommended to further analyze th e relationship.

PAGE 65

51 Figure 18 : Line diagram of aggregated Posture for each value of Heat Index.

PAGE 66

52 Figure 19 : Line diagram of aggregated Posture for each value of Heat Index by Activities. Figure 20 plots Posture vs. Activity level, colored by Activity and suggests a positive correlation exists between Posture and Activity Level.

PAGE 67

53 Figure 20 : Scatter plot showing the association between Posture and Activity Level. Next , pairwise comparisons between the different Activities were performed to investigate which activities lead to sig nificantly different Posture. Results are summarized in Table 11 .

PAGE 68

54 Table 11 : Comparison between different activities for Posture Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0 Heavy Equipment 0 0 0 Surveying 0 1 Asphalt 0 1 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwise comparisons of p values indicate that Concrete Heavy Equipment, Concrete Surveying, Concrete Asphalt, Heavy Equipment Surveying, Surveying Asphalt result in statistically distinction Posture profiles from one another. Finally, the Posture durin g Heavy Equipment is the most distinct compared to the other three Activities. Figure 21 visually confirms that Posture differs by Activity wi th Heavy E quipment having the lowest average Posture. However, it is difficult to discern which activities are statistically distinct.

PAGE 69

55 Figure 21 : Violin plot showing the distribution of Posture across the four activiti es. To understand the relationship of Posture with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Posture for each volunteer with the width of the lines showing the amount of Activity Level (Figure 22 ). This figure reveals how the Posture varies based on the choice of Activity for different volunteers. As expected, the lines become wider towards the top of the figure, but that is not always the case. For example, for

PAGE 70

56 volunteer 8, we see how the average and range of Posture is much larger as compared to Heavy equipment and Surveying despite similar Activity Levels. However, there is no consistent pattern across all volunteers. This implies that there is interaction between Volunteer and Activity and including that in the model might be ideal, but that can only be achieved using a highly complicated model which is beyond the scope of this thesis. Figure 22 : Posture versus Activity by volunte ers, width of each line represents the Activity Level. Core Temperature: Figure 23 shows the Core Temperature distribution s for the 10 volunteers. Different Volunteer s have different Core Temperature profiles . As shown, s ome Volunteer have higher mean Core Temperatures than others.

PAGE 71

57 Figure 23 : Violin plot for Core Temperature of all volunteers. To test for statistical differences a t test was performed for the Core Temperature data across all volunteers as shown Table 12 . The p values indicate that Heat Index, Activity Level and Activity are significantly associated wit h Core Temperature (p 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( ) and the residual variance ( ) are estimated to be 0.019 and 0.073 . Figure 24 Figure 27 show the associations between Core Temperature and the independent variables. Figure 24 shows the distribution of Core Temperature by Gender. It visually confirms the results of Table 12 that Core Temperature differences are not statistically significant across gender .

PAGE 72

58 Table 12 : Effect of Gender, Heat Index, Activity Level and Activity on Core Temperature Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 0.095 t=0.943 1 Heat Index Average increase in the HR when HI increases by one unit. 0.004 t=5.777 Activity level Average increase in the HR when AL increases by one unit. 0.420 t=12.531 0 Activity Average difference in HR for activities HE & C 0.0178 =1660.6 Average difference in HR for activities S & C 0.191 Average difference in HR for activities A & C 0.163 Note s : Fit of LMM for Core Temperature showing the effect of Gender, Heat Index, Activity Level and Activity on Core Temperature . The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated usin g likelihood ratio test. Y and assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the difference /increase is 0.

PAGE 73

59 Figure 24 : Violin plot showing the distribution of overall Core Temperature for Female and M ale volunteers . Figure 25 and Figure 26 depict the relationship of Core Temperature to Heat Index with each colored line representing an individual volunteer. Although the p values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Core Temperature and Heat Index. While a general trend m ay be that Core Temperature declines as Heat Index rises, it is not the case for all volunteers. M ore research is recommended to further analyze this relationship.

PAGE 74

60 Figure 25 : Line diagram of aggregated Core Temperature for e ach value of Heat Index. Figure 26 : Line diagram of aggregated Core Temperature for each value of Heat Index by Activities.

PAGE 75

61 Figure 27 plots Core Temperature vs. Activity level, colored by Activity and suggests a positive correlation exists between Core Temperature and Activity L evel. Figure 27 : Scatter plot showing the association between Core Temperature and Activity Level. Next , pairwise comparisons between the different Activities were performed to investigate which activities lead to sig nificantly different Core Temperature . Results are summarized in Table 13 .

PAGE 76

62 Table 13 : Comparison between different activities for Core Temperature Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0.324 0 0 Heavy Equipment 0.324 0 0 Surveying 0 0 0.032 Asphalt 0 0 0.032 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwise comparisons of p values indicate that Concrete Surveying, Concrete Asphalt, Heavy Equipment Surveying, Heavy Equipment Asphalt, Surveying Asphalt result in statistically distinction Heart Rate profiles from one another . Finally, the Core Temperature during S urveying and Asphalt is the most distinct . Figure 28 visually confirms that Core Temperature differs by Activity with Surveying having the lowest average Core Temperature .

PAGE 77

63 Figure 28 : Violin plot showing the distribution of Core Temperature across the four activities. To understand the relationship of Core Temperature with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Core Temperature for each volunteer with the width of the lines showing the amount of Activity Level ( Figure 29 ). This figure reveals how the Core Temperature varies based on the choice of Activity for different volunteers. We observe some difference in average Core temperature as well as the range of Core Temperatures based on activities but there is no consistent pattern across all

PAGE 78

64 volunteers. This implies that there is interaction between volunteer and Activity and including that in the model might be ideal, but that will require a highly complicated model which implies more research is recommended to analyze these relationships . Figure 29 : Core Temperature versus Activity by volunteers, width of each line represents the Activity Level. Physiological Load: Figure 30 shows the Physiological Load distribution s for the 10 volunteers. Different Volunteer s have different Physiological Load profiles. As shown, s ome Volunteer have higher mean Physiological Loads than others. The maximum buildup of Physiological Load for Volunteer 1, 3 and 8 is higher compared to the others.

PAGE 79

65 Figure 30 : Violin plot for Physiological Load of all volunteers. To tes t for statistical differences t test was performed for the Physiological Load data across all volunteers as shown Table 14 . The p values indicate that Heat Index, Activity Level and Activity are significantly associated wit h Physiological Load (p 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( ) and the residual variance ( ) are estimated to be 1.36 and 2.39. Figure 31 Figure 37 show the associations between Physiological Load and the independent vari ables. Figure 31 shows the distribution of Physiological Load by Gender.

PAGE 80

66 Table 14 : Effect of Gender, Heat Index, Activity Level and Activity on Physiological Load Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 0.947 =1.077 1 Heat Index Average increase in the HR when HI increases by one unit. 0.039 =28.603 Activity level Average increase in the HR when AL increases by one unit. 11.808 =2108.35 0 Activity Average difference in HR for activities HE & C 0.812 =591.631 Average difference in HR for activities S & C 1.359 Average difference in HR for activities A & C 0.966 Note s : Fit of CPG LMM for Physiological Load showing the effect of Gender, Heat Index, Activity Level and Activity on Physiological Load . The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated using likelihood ratio test. Y and assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the differ ence/increase is 0. All p values are conducted using likelihood ratio test.

PAGE 81

67 Figure 31 : Violin plot showing the distribution of Physiological Load for Female and M ale volunteers. Figure 32 and Figure 33 depict the relationship of Physiological Load to Heat Index with each colored line represe nting an individual volunteer. Although the p values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Physiological Load and Heat Index. While a general trend may be that Physiological Load declines as H eat Index rises, more research is recommended to further analyze this relationship.

PAGE 82

68 Figure 34 : Line diagram of aggregated Physiological Load in log scale for each value of Heat Index . Figure 35 : Line diagram of aggregated Physiological Load for each value of Heat Index.

PAGE 83

69 Figure 36 plots Physiological Load vs. Activity level, colored by Activity and suggests a positive correlation exists between Physiological Load and Activity Level. Figure 36 : Scatter plot showing the association between Physiological Load and Activity Level.

PAGE 84

70 Next , pairwise comparisons between the different Activities were performed to investigate which activities lead to sig nificantly different Physiological Loads . Resul ts are summarized in Table 15 . Table 15 : Comparison between different activities for Physiological Load Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0 0.0032 Heavy Equipment 0 0 0.054 Surveying 0.0032 0 0 Asphalt 0.054 0.003709 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwis e comparisons p values indicate that all activities result in statistically d istinction Physiological Load profiles from one another . Figure 37 visually confirms that Physiological Load differs by Activity. However, it is difficult to discern which activities are statistically distinct.

PAGE 85

71 Figure 37 : Scatter plot showing the association between Physiological Load and Activity Level. To understand the relationship of Physiological Load with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Physiological Load for each volunteer with the width of the lines showing the amount of Activity Level ( Figure 38 ). This figure reveals how the Physiological Load varies based on the choice of Activity for different volunteers. The overall Physiological Load bu ild up for activity Concrete dominates other activities ( Figure 38 ), which is also visible in Figure 37 . This implies that there is interaction

PAGE 86

72 between volunteer and Activity and including that in the model might be ideal, but that will require a highly complicated model which implies more research is recommended to analyze these relationships. Physiological Load also has a lot of zeros in the data which restricts us to obtain dependable results and conclusion. Figure 38 : Physiological Load versu s Activity by volunteers , width of each line represents the Activity Level . Mechanical L oad: Figure 39 shows the Mechanical Load distributions for the 10 volunteers. Different Volunteers have different profiles in Mechanical Load and it is difficult to observe based on the scale of the graph. However, it is possible to observe that some Volunteer have higher mean Mechanical Loads than others.

PAGE 87

73 Figure 39 : Violin plot for Mechanical Load of all volunteers. To test for statistical differences a t test was performed for the Mechanical Load data across all volunteers . Table 16 shows the fitted CPG LMM and the p values corresponding to the hypotheses to be tested for Q1, Q2, and Q3 . The p values indicate that Heat Index, Activity Level and Activity are significantly associated wit h Mechanical Load (p 0.05), but gender is not significantly associated (p > 0.0 5). The volunteer specific variance ( ) and the residual variance ( ) are estimate d to be 0.10 and 0.85. Figure 40 Figure 43 show the associations between Mechanical Load and the independent variables. Figure 40 shows the distribution of Mechanical Load by Gender .

PAGE 88

74 Table 16 : Effect of Gender, Heat Index, Activity Level and Activity on Mechanical Load Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 0.169 =0.472 1 Heat Index Average increase in the HR when HI increases by one unit. 0.007 =1.850 1 Activity level Average increase in the HR when AL increases by one unit. 15.401 =6184.8 45 0 Activity Average difference in HR for activities HE & C 0.834 =544.32 5 Average difference in HR for activities S & C 0.277 Average difference in HR for activities A & C 0.040 Note s : Fit of CPG LMM for Mechanical Load showing the effect of Gender, Heat Index, Activity Level and Activity on Mechanical Load . The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated using likelihood ratio test. Y and assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the difference/increase is 0 . All p va lues are conducted using likelihood ratio test.

PAGE 89

75 Figure 40 : Violin plot showing the distribution of Mechanical Load for Female and M ale volunteers. Figure 41 and Figure 42 depict th e relationship of Mechanical Load to Heat Index with each colo red line representing an individual volunteer. Although the p values are significant, it is difficult to draw clear conclusions about patterns in the relationship between Mechanical Load and Heat Index. While a general trend may be that Mechanical Load dec lines as Heat Index rises, more research is recommended to further analyze this relationship.

PAGE 90

76 Figure 41 : Line diagram of aggregated Mechanical Load in log scale for each value of Heat Index. Figure 42 : Line diagram of aggregated Mechanical Load for each value of Heat Index.

PAGE 91

77 Figure 43 plots Mechanical Load vs. Activity level, colored by Activity and suggests a low positive correlation exists between Heart Rate and Activity Level. Figure 43 : Scatter plot showing the association between Mechanical Load and Activity Level. Next, pairwise comparisons between the different Activities were performed to investigate which activities lead to sig nificantly different Mechanical Load . Results are summ arized in Table 17 .

PAGE 92

78 Table 17 : Comparison between different activities for Mechanical Load Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0.0001 1 Heavy Equipment Surveying 0.0001 1 Asphalt 1 1 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. The pairwise comparisons of p values indicate that Concrete Heavy Equipment, Concrete Surveying, Heavy Equipment Surveying, Heavy Equipment Asphalt result in statistically distinction Mechanical Load profiles from one another. . Finally, the Mechanical Load during Heavy Equipment is the most distinct comp ared to the other three Activities.

PAGE 93

79 Figure 44 : Violin plot showing the distribution of Mechanical Load across the four activities. Figure 44 v isually con firms that Mechanical Load differs by Activity with Surveying havi ng the lowest a verage Mechanical Load . To understand the relationship of Mechanical Load with the Activities also accounting for the difference between volunteers and the difference in Activity Level, we used a plot of Activity versus Mechanical Load for each volunteer with the width of the lines showing the amount of Activity Level ( Figure 45 ). This figure reveals how the Mechanical Load varies based on the choice of Activity for different volunteers.

PAGE 94

80 However, due to numerous zero values in the data, we cannot conclude anything else from Mechanical Load . M ore research is recommended to analyze these relationships . Figure 45 : Mechanical Load versus Activity by volunteers , width of each line represents the Activity Level. Activity Level: Figure 46 shows the Heart rate distributions for the 10 volunteers. Different Volunteers have different profiles for Activity Level . As shown, some Volunteer have higher mean Activity Level than others. For example volunteer 6 and 7 seemed to have the highest mean s as compared to the others.

PAGE 95

81 Figure 46 : Violin plot for Activity Level in log scale of all volunteers. To test for statistical differences a T test was performed for the Activity Level data across all volunteers as shown Table 18 . Table 18 : Effect of Ge nder, Heat Index and Activity on Activity Leve l Effect Coefficient Interpretation Estimate Test statistic P value Gender Average difference in HR between males and females. 0.364 t= 3.374 0.630 Heat Index Average increase in the HR when HI increases by one unit. 0.008 t= 4.068 0.003 Activity Average difference in HR for activities HE & C 0.077 =316.46 Average difference in HR for activities S & C 0.295 Average difference in HR for activities A & C 0.200 Note s : Fit of LMM for Activity Level showing the effect of Gender, Heat Index, Activity Level and Activity on Activity Level . The p values for Gender. Heat Index and Activity Level are calculated using t test and the p value for Activity was calculated using likelihood ratio test.

PAGE 96

82 means X Y and assumes that the values of all other variables remain the same. The test statistic and the p values are related to the hypothesis that the difference/increase is 0. The p values indicate that Heat Index and Activity are significantly associated with Activity Level (p 0.05), but gender is not significantly associated (p > 0.05). The volunteer specific variance ( ) and the residual variance ( ) are estimated to be 0.01852 and 0.66264. Figure 47 Figure 49 show the associations between Activity Level and the independent variable s. Figure 47 shows the distribution of Activity Level by Gender. It visually confirms the results of Table 18 that Activity Level differences are not statistically significant across gender.

PAGE 97

83 Figure 47 : Violin plot showing the distribution of overall Activity Level in log scale for Female and M ale volunteers. Figure 48 and Figure 49 depict the relationship of Activity Level to Heat Index with each colored line representing an individual volunteer. Although the p values are significant, it is difficult to draw clear conc lusions about patterns in the relationship between Activity Level and Heat Index. While a general trend may be that Activity Level declines as Heat Index rises , there are a few exceptions , and more research is recommended to further analyze this relationsh ip. Similarly as explained for Heart Rate earlier, the change of Activity Level will be negligible for a unit change in Heat Index.

PAGE 98

84 Since the number of observations is very large, the p values can be very small even when there is a small, pra ctically negli gible effect size. Figure 48 : Line diagram of aggregated Activity Level in log scale for each value of Heat Index.

PAGE 99

85 Figure 49 : Line diagram of aggregated Activity Level in log scale for each value of Heat Index by Activities. Next , pairwise comparisons between the different Activities were performed to investigate which activities lead to significantly different Activity Level . Results are summarized in Table 19 . Table 19 : Comparison between different activities for Activity Level Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0.002 0 0 Heavy Equipment 0.002 0 Surveying 0 0 0.003 Asphalt 0 0.003 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison.

PAGE 100

86 The pairwise comparisons p values indicated that all activities result in statistically distinction Activity Level profiles from one another. Figure 50 : Violin plot showing the distribution of Activity Level in log scale across the four activities. Figure 50 visually confirms that Activity Level differs by Activity with Surveying having the lowest average Activity Level , and Concrete Placement having the highest . However, it is difficult to discern which activities are statistically distinct.

PAGE 101

87 I llustration of the effects using four chosen volunteers : After doing analysis based on metrics across all volunteers, the author analyzed data from four volunteers from the same flight. Volunteer 1, 2, 3 and 4 were from same flight , and were, therefore, working under same climatic conditions and doing the same construction activities synchronously . Figure 51 Figure 58 shows the violin plots for each metric against the volunteer s for all act ivities. Figure 51 : Violin plot showing the distribution of Heart Rate for volunteer 1,2,3,4 across the four activities. Figure 52 : Violin plot showing the distribution of Breathing Rate for volunteer 1,2,3,4 across the four activities. Figure 53 : Violin plot showing the distribution of Core Temperature for volunteer 1,2,3,4 across the four activities.

PAGE 102

88 Figure 54 : Violin plot showing th e distribution of Posture for volunteer 1,2,3,4 across the four activities (taking absolute value for Posture). Figure 55 : Violin plot showing the distribution of Posture for volunteer 1,2,3,4 across the four activities. Figure 56 : Violin plot showing the distribution of Activity Level for volunteer 1,2,3,4 across the four activities.

PAGE 103

89 Figure 57 : Violin plot showing the distribution of Mechanical Load for volunteer 1,2, 3,4 across the four activities. Figure 58 : Violin plot showing the distribution of Physiological Load for volunteer 1,2,3,4 across the four activities. I n Figure 51 when Heart Rate is compared for the four volunteers, there is a clear pattern showing that Heart Rate for volunteer 1 and 4 are always greater , on average, compared to volunteer 2 and 3 . We c an conclude that Heart rate is in general higher for volunteer 1 and 2 irrespective of activity type. However, this whole pattern is shifted up or down based on the Activity. For example, the Heart Rates of all four volunteers are lower, on average, for Su rveying. Also, the Heart Rates reach a much higher value for Concrete compared to other activities. Similar pattern s are noticed for Breathing Rate in Figure 52 and Core Temperature in Figure 53.

PAGE 104

90 In Figure 55 , posture is plotted for all the four volunteers across the four activities. It is interesting to notice that volunteer 2 has more negative values as c ompared to the other volunteers . Posture is measured as the change of angle of any individual in comparison with the gravity. When any individual is standing straight the measurement is zero. Forward and backward leaning accounts for positive and negative values. The negative value indicated how the particular vo lunteer has leaned back more for all the activities. Figure 57 and Figure 58 are the violin plots for the Mechanical Load and Physiological Load for the four volunteers across the same four activities. It is difficult to understand any trend from t hese two figures since the data for Mechanical Load and Physiological Load contains many zero values . We then followed up and conducted the pairwise comparisons between the different Activities to investigate which activities lead to significantly different Heart Rate for volunteer 1 through volunteer 4. Volunteer 1, 2, 3, 4 were from the same flight and h ence were invol ved in similar activities and exposed to same weather conditions. Table 20 shows t he pairwise comparisons p values indicated that all pairs of ac tivities except Asphalt and Concrete yi eld significantly different Heart Rates. Table 21 shows that Breathing Rate is similar for Heavy Equipment, Concrete and Asphalt. Table 22 shows Core Temperature is significantly different for all activities.

PAGE 105

91 Table 23 we can conclude that Asphalt Concrete and Asphalt Surveying have simi lar posture and Heavy equipment yields most significant Posture. Table 24 shows that Surveying yields significantly different Activity Level. Table 25 shows that Mechanical Load is statistically significant for all the four activities. Table 26 shows that Physiological Load is statistically significan t for all the four activities. Table 20 : Comparison between different a ctivities for Heart Rate Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0 1 Heavy Equipment 0 0 0 Surveying 0 0 0 Asphalt 1 0 0 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. Table 21 : Comparison between different activities for Breathing Rate Activity Concrete Heavy Equipment Surveying Asphalt Concrete 1 1 Heavy Equipment 1 0 0 Surveying 0 0.0009 Asphalt 1 1 0.0009 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison.

PAGE 106

92 Table 22 : Comparison between different activities for Core Temperature Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0.006 0 0 Heavy Equipment 0.006 0 0 Surveying 0 0 Asphalt 0 0 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. Table 23 : Comparison between different activities for Posture Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0 1 Heavy Equipment 0 Surveying 1 Asphalt 1 1 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. Table 24 : Comparison between different activities for Activity Level Activity Concrete Heavy Equipment Surveying Asphalt Concrete 1 0 0.176 Heavy Equipment 1 0 0.668 Surveying 0 Asphalt 0.176 0.668 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison.

PAGE 107

93 Table 25 : Comparison between different activities for Mechanical Load Activity Concrete Heavy Equipment Surveying Asphalt Concrete 0.045 Heavy Equipment 0.838 Surveying Asphalt 0.045 0.838 Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison. Table 26 : Comparison between different activities for Physiological Load Activity Concrete Heavy Equipment Surveying Asphalt Concrete Heavy Equipment Surveying Asphalt Note s : Pairwise follow up comparison between different activities. Each cell in the table shows the Bonferroni adjusted p value for the corresponding pairwise comparison.

PAGE 108

94 CHAPTER 4: CONCLUSION Summary and Discussion This research analyzed a pilot dataset to explore statistical relationships and observable patters for physiological metrics variables as measured for construction workers performing four distinct construction activities. Table 27 provides a summary of the analysis for all metrics. The underlying contribution is to describe and analyze physiological data for construction workers to assess health and productivity impacts of perform ing discrete construction tasks. Table 27 : Summary Ta ble of all metrics and their association (p values) with the explanatory variables Metrics Gender Heat index Activity Level Activity Heart Rate 1 0.033 0 6.6234 × 10 65 Breathing Rate 1 5.32 × 10 7 0 0 Core Temp 1 0 0 1.0677 × 10 83 Posture 1 0 0 1.0677 × 10 83 Activity Level 0.630 0.003 1.88 × 10 66 Mechanical load 1 1 0 8.14 × Physiological Load 1 6.13 × 10 6 0 4.53 ×

PAGE 109

95 G ender is not statistically significant for any of the metrics. Heat Index, in most cases, has a significant P observable in graphical representations. This implies that the eff ect sizes are not large. It is interesting to note that Activity Level and Activity each have a positive correlati on with most of the metrics , and hence, may be an important factor to predict the health conditions of any construction worker. Activity can potentially mediate the effect of Activity on the other metrics in the sense that certain Activity may lead to high er Activity Level which may in turn influence the other metrics. From the illustrations, it is evident that such mediation is partial, i.e. for most metrics, there is a direct effect of Activity , and a part of the effect of Activity is perhaps mediated thr ough Activity Level. A better understanding of such mediation requires more sophisticated statistical models and can be the subject of future studies. In general, a relationship of volunteers and Activity is observable. Further study using more sophisticated models may reveal how the interaction of Activity Level and Activity effects the volunteer. This prediction can be used as a precautionary measure against occupational health hazard in any jobsite to avoid accidents. Eventually, it may also be possible to understand what kind of physical conditions lead the al metrics to exceed a threshold li mit. Using the prediction model, it may be possible, to prevent such situations on a jobsite.

PAGE 110

96 Results ( Figure 7 , Figure 14 , Figure 28 , and Figure 21 ) also indicate that Concrete Placement is most demanding for all volunteers. A more in depth look at the four volunt eers performing construction activities concurrently (Figure 51 Figure 56) , also shows that Concrete Placement results in higher metric measures than other activities . Finally, there is a general trend, as seen in Figure 51, that the mean of Heart rate f or volunteers 2 and 3 is greater than volunteer 1 and 4, across all activities. Figure 54 , shows that the mean of the posture metric for volunteer 1 and 4 is greater than volunteer 2 and 3 across all activities. This shows how the metrics are independent of each other and effect all volunteers differently. In general, results of the study suggest that heat index and construction activities affect the physiological metrics of the individual cadets differently, despite individuals having relatively similar physical characteristics. Furthermore, results suggest that concrete and asphalt placement are generally the most physically demanding construction activity studied, followed by, heavy equipment and surveying activities respectively. Finally, the study shows that it is possible to independently compare discrete physiological metrics across individuals as well as activities. The research serves to highlight significant opportunity to use such methods to study construction worker health and productivity in the future.

PAGE 111

97 Delimitations This study does not account for the accuracy of the measurements using the bioharness. Any error in the measuring devices can lead to biased results, and although outlying data wer e filtered, accuracy was not validated . For example, many zeros in the Physiological Load and Mechanical Load data can raise some concern about the accuracy of those measures. Our statistical analysis also appears to be over powered due to very large amoun t of data resulting in very small p values which can lead to the misconception that every variable has a very strong influence. However, we have to remember that p values is only a measure of how strongly we believe that there is an effect. It does not pro vide the magnitude of the effect. Looking at the effect sizes (as provided in the table) and the visualization of the data besides considering the p values can lead us to a more meaningful conclusion. Future studies using more sophisticated methods, e.g. t ime series analysis, may resolve this pro blem of over powered analysis. Once the Zephyr Bioharness System were purchased, we did not calibrate them by ourselves. We assumed they were pre calibrated. Future Research: There are some limitations of our analysis. As we have mentioned earlier, due to the large amount of data, the statistical method tends to detect very small effect sizes as significant. Looking at the effect size besides the p value is recommended. This can also be fixed using more sophisticated statistical methods.

PAGE 112

98 Secondly, we have many zeros for the P hysiological Load and M echanical L oad data which prevented us from having a good visual understanding of the data . Better data can provide better insight of the load build up. We did not specifically model interactions, which can be present. More complicated statistical methods can be used to model the interactions between activities and volunteers.

PAGE 113

99 References : Journal of Construction Engineering and Management , 128(5), 427 437. Abudayyeh, O., Fredericks, T. K., International Journal of Project Management , 24(2), 167 174. Aryal, A., Ghahramani, A., and Becerik construct Automation in Construction , Elsevier B.V., 82, 154 165. personalized construction safety monitoring and trending: Review of applicable Automation in Construction , Elsevier, 85(July 2016), 96 106. Banos, O., Villalonga, C., Damas, M., Gloesekoetter, P., Pomares, H., and Rojas, I. a Ubiquitous, Continu Scientific World Journal , 2014. Journal of Occupational Medici ne and Toxicology , 3(1), 1 10.

PAGE 114

100 environment: comparison of accelerometry and oxygen consumption/heart rate Ergonomics , Taylor & Francis, 51(5), 663 670. Buller , M. J. . B., Tharion, W. J. ., Hoyt, R. W. ., and Chadwicke Jenkins, O. . (2010). Proceedings of the National Conference on Artificial Intelligence , 3, 1763 1768. Buller, M. J 1768. Time Location Sensing and Physiological Status Monitoring for Ergonomics Analysis Journal of Computing in Civil Engineering , 27(3), 320 335. C time resource location data collection and visualization technology for construction safety and activity monitoring Automation in Construction , Elsevier B.V., 34, 3 15. specific temporal Automation in Construction , 84(August 2016), 335 344. acceptance of wearable technologies in the workplace?: Indoor localization and Automation in Construction , Elsevier, 84(July 2016), 31 41.

PAGE 115

101 Field Engineering Cadet Journal of the American Statistical Association , 56(293), 52 64. based fall portent detection algorithm for construction tilin Automation in Construction , Elsevier, 84(July 2016), 214 230. Journal of occupational and environmental hygiene , Taylor & Fra ncis Group, 2(4), 203 212. Garet, M., Boudet, G., Coudert, J., Montaurier, C., Vermorel, M., and Chamoux, A. validation by whole European Journal of A pplied Physiology , 94(1 2), 46 53. study of the relationship between construction workforce physical strain and task Construction Management and E conomics , Routledge, 32(6), 548 564. American journal of industrial medicine , Wiley Online Library, 42(5), 384 396. Griffin, W. L., Pe arson, N. J., Belousova, E., Jackson, S. E. v, Van Achterbergh, E.,

PAGE 116

102 mantle: LAM MC Geochimica et Cosmochimica Acta , Elsevi er, 64(1), 133 147. Automation in Construction , Elsevier B.V., 82, 138 153. si Scandinavian journal of work, environment & health , JSTOR, 88 95. American journal of industrial medicine , Wiley Onl ine Library, 44(4), 377 384. type wearable health devices to Automation in Construction , Elsevier, 83(August 2016), 330 340. Kim, J. H., Roberge, R., Powell, J. B ., Shafer, A. B., and Williams, W. J. (2013). and sustained exercise in the heat using the Zephyr BioHarness TM International journal of sports medicine , NIH Public Access, 34 (6), 497. setters in New Zealand logging operations Heart rate strain in cable hauler choker 5. Kirk,

PAGE 117

103 Applied Ergonomics , 32(4), 389 398. monitoring on duty and off duty worker physiological status and activities in Automation in Construction , Elsevier, 83(August 2016), 341 353. Lee, W., and Mig Construction Innovation , 16(3), 281 306. 2011 Copyright ASCE 2011 Computing in Civil Engine Civil Engineering , (413), 760 767. 548. B., Heller, Arthritis & Rheumatism: Official Journal of the American College of Rheumatology , Wiley Onli ne Library, 43(5), 1021 1032. muscle fatigue of road maintenance workers based on heart rate monitoring and Journal of occupational medicine and toxic ology (London, England) , 1, 20.

PAGE 118

104 to Know About Heat 2. climatic heat stress risk in construction: A review of practices, methodologies, Accident Analysis and Prevention , 66, 187 198. Sandmark, Occupational and environmental medicine , BMJ Publishing Group Ltd, 57(10), 673 677. The Sixth Annual International Conference on Computing in Civil and Building Engineering , (1967), 455 462. usculoskeletal injuries in construction: a review of the Applied occupational and environmental hygiene , Taylor & Francis Group, 16(11), 1056 1064. potential hazards American Industrial Hygiene Association Journal , Taylor & Francis, 55(7), 635 649. Tak, S. W., Buchholz, B., Punnett, L., Moir, S., Paquet, V., Fulmer, S., Marucci ghway tunnel construction: Overview from the Construction Occupational Health

PAGE 119

105 Applied Ergonomics , Elsevier Ltd, 42(5), 665 671. Taylor, P., Merlino, L. A., Rosecrance, J. C., Anton, D., Cook, T. M., Merlino, L. A., Rosecrance, J. C., Anton, D., a Environmental Hygiene Symptoms of Musculoskeletal Disorders Among Apprentice Construction Workers Symptoms of Musculoskeletal Disorders Among Apprentice 41. Tei Automation in Construction , Elsevier B.V., 35, 53 68. Wang, D., Chen, J., Automation in Construction , Elsevier B.V., 82, 122 137. Wang, Z., and Journal of Occupational and Environmental Hygiene , Taylor & Francis, 13(3), 185 192. Automation in Construction , Elsevier, 82(August 2016), 166 178. Journal of Computing in Civil Engineering , 29(5), 05014009.

PAGE 120

106 Automation in Construction , Elsevier B.V ., 62, 101 113. real Automation in Construction , Elsevier, 82(July 2016), 193 206. Zephyr. (2013). Zephyr Technologies . 18. cons truction sites using Internet of Automation in Construction , Elsevier, 83(May), 372 389.