Stress Monitoring in Construction (Fall 2024) - CivE709B, PDF

Fall 2024, CivE709B Construction Safety and Human Factors 7. Stress Monitoring in Construction - Potential & Barriers of Wearables - Gaang Lee, PhD. Assistant Professor...

Fall 2024, CivE709B Construction Safety and Human Factors 7. Stress Monitoring in Construction - Potential & Barriers of Wearables - Gaang Lee, PhD. Assistant Professor Hole School of Construction Dept. of Civil & Environmental Engineering University of Alberta A U TTENTIVE  How to Manage Stress? Monitoring is First! I will not cover the interventions & management strategies - You can easily search! I will focus on “stress monitoring” - Q. Why is field stress monitoring crucial in managing worker stress, even when we already know the typical stressors in construction? A U TTENTIVE 2  Another Reason Why We need to Monitor Stress Not just managing stress Managing overall quality of interactions b/w workers & sites in a proactive manner Stress is a key indicator in managing not just safety, but also, health, productivity, and satisfaction - Indication of overall quality of interactions w/ environments, regardless of the nature of the issue - Leading indicators of accidents / illness → proactive interventions - Stress → unsafe behavior → accidents - Stress → prolonged → illnesses A U TTENTIVE 3  Let’s Study How to Monitor Stress in the Field Management Stress Management A U TTENTIVE 4  Outline Traditional monitoring techniques & limitations Wearable biosensors’ potential Advantages Different biosensors & pros and cons Wearable biosensor + machine learning for field stress monitoring How they work together Limitations Barriers to applying wearable biosensors in the field A U TTENTIVE 5  Traditional Stress Monitoring Techniques Surveys - Questionnaires - Perceived Stress Scale (PSS) (Cohen et al. 1983) “For the 30 days,” Overall - Job Stress Survey (JSS) (Parker and DeCotiis 1983) Specific to the workplace - Stress Appraisal Measure (SAM) (Peacock and Wong 1990) Sensitive to threat/challenge - Cognitive Appraisal Ratio (Feinberg and Aiello 2010) Very short version of SAM - Interviews & manual observations A U TTENTIVE 6  Management Stress Management Sporadic Invasive Questionnaires Interviews Observations A U TTENTIVE 7  Traditional Stress Monitoring Techniques Surveys - Questionnaires - Perceived Stress Scale (PSS) (Cohen et al. 1983) “For the 30 days,” Overall - Job Stress Survey (JSS) (Parker and DeCotiis 1983) Specific to the workplace - Stress Appraisal Measure (SAM) (Peacock and Wong 1990) Sensitive to threat/challenge - Cognitive Appraisal Ratio (Feinberg and Aiello 2010) Very short version of SAM - Interviews & manual observations Limitations - Invasive, requiring workers’ time commitment and active participation - Thus, should be sporadic and post-hoc manners, not continuous - Thus, not matched with the construction industry’s dynamic nature - Thus, not good for the proactive interventions A U TTENTIVE 8  Outline Traditional monitoring techniques & limitations Wearable biosensors’ potential Advantages Different biosensors & pros and cons Wearable biosensor + machine learning for field stress monitoring How they work together Limitations Barriers to applying wearable biosensors in the field A U TTENTIVE 9  Wearable Biosensors’ Potentials Not field-applied yet Stressor Central Response SAM/HPA Axes Peripheral Reactivity Biosignals from Peripheral ECG, PPG, EDA, ST, Eye-tracking Biosignals from Central Advantages - Non-Invasive, non-interfering with work - Thus, can be continuous & capture dynamics - Thus, good for the proactive interventions EEG, fNIRS A U TTENTIVE 10  Different Biosignals Collected by Wearables Biosignals coming from peripheral systems - ECG (electrocardiogram) - PPG (photoplethysmogram) - EDA (eletrodermal activity; a.k.a, galvanic skin response) - ST (skin temperature) - Eye-tracking Biosignals coming from central systems - EEG (electroencephalogram) - fNIRS (Functional Near-Infrared Spectroscopy) A U TTENTIVE 11  Electrocardiogram (ECG) Using electrodes, recording the “heart” electrical activity - SAM, sympathetic arousal → heart activities (cardiovascular activities) Metrics - Heart rate (HR: how many times the heart beats per minute (bpm)) - Normal under rest: 50-70bpm & stress increases it - Heart rate variability metrics (HRV; variability of HR) – time domain - SDNN (Standard deviation of NN intervals), stress decreases it - RMSSD (Root Mean Square of Successive Differences), stress decreases it A U TTENTIVE 12  Time Domain vs. Frequency Domain Conversion w/ Furrier Transform - Time Domain: x axis is time (typically what we see from a graph) - Frequency Domain: x axis is frequency (we get it by furrier-transforming a signal fragment A U TTENTIVE 13  Electrocardiogram (ECG) (Cont.) Metrics - Heart rate variability metrics (HRV; variability of HR) – frequency domain - Low-Frequency Power (LF): Reflects sympathetic activity; stress increases it - High-Frequency Power (HF): Reflects parasympathetic activity; stress decreases it - Ratio (LF/HF): stress increases it Pros & Cons - The most accurate wearable to observe heart activity - Somehow, invasive; needs to contact skins near heart - Relying on SAM-related peripheral response Q. example? - Hard to differentiate stress from other states activating SAM - Hard to differentiate threat/challenge (SAM is activated in both) A U TTENTIVE 14  Photoplethysmogram (PPG) By shining lights onto the skin and detecting the absorbed or reflected lights, understand changes in blood volume under the skin part - SAM → heart → blood movement → blood volumetric change (so, some calls it PBV) Pulse Blood Volume Metrics: share the metrics with ECG - The peaks in PPG will indicate R peaks in ECG (with a slight delay) - So, we try to measure HR and HRV metrics from the peaks in PPG - (e.g., HR, SDNN, RMSSD, HF, LF, LF/HF) A U TTENTIVE 15  Photoplethysmogram (PPG) (Cont.) Pros - Much less invasive than ECG (so; very popular (e.g., Apple watch, Google watch, Fitbit, etc.)) - Very affordable (less than $5) - Well reflecting heart activities (correlations with ECG) Cons - Less accurate than ECG - Relying on SAM-related peripheral response (Kumar et al. 2021) - More subject to motions (than ECG) - Bring changes in the shining & reflection angle - So, there have been many techniques to control it A U TTENTIVE 16  Electrodermal Activity (EDA) Different components (Benedek and Kaernbach 2010) - EDA = EDL (ED level) + EDR (ED response) + Noise - EDL: Tonic EDA, slowing changing stress level - EDR: Phasic EDA, responses to immediate stressors (2-10 seconds delays) EDA decomposition techniques - Deconvolution-based technique (Benedek and Kaernbach 2010) - Convex optimization-based technique (Greco et al. 2016) (Amin and Faghih 2019) A U TTENTIVE 17  Electrodermal Activity (EDA) (Cont.) Metrics - Time domains: SD, mean, root mean square of EDL & EDR (stress increases) - Frequency domains: power of different band frequencies from EDR (stress increases) Pros - Less invasive than ECG - Less subject to motion artifact than PPG (but still subject somehow..) - Reasonably cheap ($15~25) - Can have two focuses in the measurement - Interactions with immediate stressors from EDR - Long-term trends from EDL A U TTENTIVE Figure: keysight.com 18  Electrodermal Activity (EDA) (Cont.) One example focusing on the response to immediate stressors (Lee et al. 2020) - Crowd-sensing environmental stressors via EDA wristband (EDR focus) GPS A U TTENTIVE 19  Electrodermal Activity (EDA) (Cont.) One example focusing on the response to immediate stressors (Lee et al. 2020) Wearable Biosensing Individual Stress Detection Stress Hotspot Identification Controlled Route Data Collection Labeled Data Signal Processing Collectively Geocoded Stress Feature Extraction Classification Comparing w/ Random Simulations Daily Trip Data Collection Cases Sig. 0.01 Simulations Actual … + Stress Density GPS Q. What can we do this hotspot analysis with wearables in construction? A U TTENTIVE 20  Electrodermal Activity (EDA) (Cont.) Cons - Still not robust enough to motion: - Motion can change quality of contact b/w skins and electrodes - Subject to ambient humidity - Lack of humidity → unstable contact b/w skins and electrodes - Excessive humidity → conductance is governed by skin moisture than gland activities - Working range: 30-60% Relative Humidity - Relying on SAM-related peripheral response - Hard to differentiate stress from other states activating SAM - Hard to differentiate threat/challenge (SAM is activated in both) A U TTENTIVE 21  Two Mobile Systems for Brain, the Central System EEG (electroencephalogram) fNIRS (Functional Near-Infrared Spectroscopy) A U TTENTIVE 22  Electroencephalogram (EEG) With electrodes, observing electric activities caused by neural activities - Simply speaking, a brain version of ECG/EDA - Making current b/w electrodes, checking voltage fluctuation generated by neural activities - Multiple electrodes → multiple channels → regional analyses Amplifier Electrode Locations EEG Signals EEG Mechanisms A U TTENTIVE 23  Electroencephalogram (EEG) (Cont.) General EEG analysis - Frequency-wise: Different frequency bands might have different indications - Delta (0.5 – 4 Hz): Deep sleep, unconsciousness, and brain healing processes - Theta (4 – 8 Hz): Drowsiness, light sleep, creativity, and deep relaxation - Alpha (8 – 12 Hz): Relaxation, calm, and wakeful rest - Beta (12 – 30 Hz): Active thinking, focus, alertness, and problem-solving - Gamma (30 – 100 Hz): High-level information processing, cognitive functioning, and consciousness - Combined with Frequency & Regions A U TTENTIVE Figure: neuphony.com 24  Electroencephalogram (EEG) (Cont.) General EEG analysis - Network-wise: each network is highly activated when: - Default Mode Network (DMN): when you’re not focused on tasks, letting your mind wander freely - Central Executive Network (CEN): when you’re focused, planning, or problem-solving - Salience Network (SN): When you focus on a silent stimuli (brain’s attention police). It helps you shift between tasks, react to important stimuli, and stay aware of your surroundings. - Microstate A: when mainly involved in perceiving and processing sound - Microstate B: when mainly involved in perceiving and processing vision - Very helpful to understand the cognitive process (Kronke et al. 2020) (Antonova et al. 2022) A U TTENTIVE Figure: neuphony.com 25  Electroencephalogram (EEG) (Cont.) Stress-related EEG analysis - Validated metrics: no one specific for stress, but there are some related - Arousal (mental activation): beta power / alpha power (Klimesch 1999) - Valence (pleasantness): frontal alpha asymmetry = alpha (frontal right) – beta (frontal lest) (Davidson 1998) - Correlations - Stress: High Arousal - Challenge: High Arousal w/ High Valence - Threat: High Arousal w/ Low Valence 2D Emotion Map (Islam et al. 2021) A U TTENTIVE Figure: neuphony.com 26  Electroencephalogram (EEG) (Cont.) Stress-related EEG analysis - Nowadays, machine learning models are developed and used Pros (in terms of stress monitoring in the field) - Can differentiate threat and challenge!; brain determines the different physiological responses Cons (in terms of stress monitoring in the field) - Invasive (honestly..), most type requires applying gel on the hair - Very subject to motion artifact (hall effect) - Very expensive (~$40,000) A U TTENTIVE Figure: neuphony.com 27