Determination of Occupational Fatigue Detection Thresholds Using Naturalistic Fatigue Datasets

Abstract

Fatigue has led to accidents, injuries, and disasters in the oil and gas industry. While subjective methods have been dominant for fatigue measurement in occupational settings, performance metrics have been used to provide additional insights into vigilance and alertness, providing an opportunity to use composite metrics to determine fatigue. This study investigated the development and validation of a composite fatigue metric that demonstrated sensitivity across various levels of fatigue and its effects on multiple outcomes. Physiological (heart rate variability), sleep (time and efficiency), subjective (Borg’s Rate of Perceived Exertion [RPE], Karolinska Sleepiness Scale [KSS]), Mental Exhaustion [ME]), and performance metrics (Psychomotor Vigilance Task [PVT] reaction times and lapses) were obtained daily for 14 days from seventy offshore workers across two oil platforms. Unsupervised machine learning methods were used to create fatigue groups, and statistical tests were used to assess the differences in fatigue outcomes between the groups. The fatigue groups using a combination of PVT lapses and ME by the DBSCAN algorithm provided the largest differences. These groups were validated with significant differences in RPE, KSS, and pre-shift heart rate variability metrics. These findings provide occupational fatigue detection thresholds using a combination of subjective and performance measures, which may aid in developing fatigue models that capture physiological changes related to fatigue.