Date of Award


Document Type


Degree Name

Master of Science (MS)


Biomedical Engineering

First Advisor

Kouhyar Tavakolian


Drowsiness is a transitional psychophysiological state from alertness towards sleep, which decreases concentration and increases response time. Drowsiness during duty hours is common for in-flight pilots due to frequent travel across different time zones, extended duty hours as well as circadian rhythm disruption. Hence, drowsy flying is one of the leading reasons for increased risk of accidents, especially in commercial aviation. Mainly three approaches (i.e., vehicle-based, behavioral, and physiological signal based) are used for onboard drowsiness detection. Among them, physiological signal-based approach is advantageous for early detection of drowsiness with reasonable accuracy due to the strong relationship among some of the physiological signals (e.g., cardiac signal, brain wave) and psychophysiological states. Continuous monitoring of these physiological signals can be useful for early drowsiness detection. In this study of pilots’ drowsiness detection, potentials of Electroencephalogram (EEG), Electrocardiogram (ECG), and Photoplethysmogram (PPG) have been explored for on-board wearable drowsiness detection and warning system design. ECG, ear PPG, EEG, and vertical Electrooculogram (EOG) were recorded from 18 commercially rated pilots from 02:00 AM to 04:30 AM during simulated flight operation. In the case of EEG analysis, power spectral density (PSD) estimation has been used. Relative band power changes during microsleep (MS, <15s) and longesleep (LS, >15s) periods compared to baseline periods were tested for four EEG frequency bands (delta (δ, 0.5-4Hz), theta (θ, 4-8Hz), alpha (α, 8-13Hz), and beta (β, 13-30Hz)) from five brain regions ((Frontal, F), (Central, C), (Parietal, P), (Temporal, T), and (Occipital, O)). Delta band power reduced significantly (p<0.05) during microsleep periods, whereas alpha band power showed a significant increase during microsleep events for all the brain regions. Theta and beta band power did not show any significant xv change during drowsiness. RR intervals using ECG, PP intervals, crest time, diastolic peak time, systolic peak to diastolic peak, and diastolic time using PPG increased significantly during drowsy periods. Pulse arrival time (PAT) calculated using ECG R-peak and PPG peak increased significantly (p<0.05) during drowsy periods compared to baseline (443.51±14.07ms vs. 407.66±09.85ms). However, decrease in PAT/RR during drowsy periods for most of the subjects indicates that increase in PAT and RR intervals during drowsy periods are not linearly correlated; and PAT/RR can be used as an independent feature for drowsiness estimation. Besides, some other heart rate variability (HRV) features (e.g., SDNN, RMSSD, LF, and HF) showed significant change during drowsy periods. This study shows that besides mostly used EEG, which is not quite feasible for on-board applications due to the requirement of numerous electrodes placement on the scalp, both ECG and PPG can be used to monitor the physiological changes during drowsy periods. Especially, PPG has the potential for wearable applications, since it is easily obtainable compared to both EEG and ECG. However, studies on more subjects with variations in age range, different parts of the day, and study environment are required for generalizing current findings and universal recommendations.


This thesis was updated with minor corrections on July 24, 2020. For access to the version available prior to this date please contact the Digital Initiatives Librarian.