Fatigue Detection in Commercial Flight Operations: Results Using Physiological Measures☆

In an effort to better understand the impact of fatigue on commercial flight operations, we collected physiological and performance data from commercial flight crews performing simulated operations under both rested and fatigued conditions. The purpose of this research was 1) to evaluate the effects of varying levels of fatigue and workload on pilot performance and physiological responses, and 2) to determine whether any technology exists to detect symptoms of fatigue in real time. Thirty-two airline pilots were fitted with a variety of physiological measurement devices to measure characteristics that are affected by fatigue. Each crew of two pilots flew a “rested” and a “fatigued” session in a high-fidelity B-777 simulator. Crews were assigned to one of two flight scenarios: either one 6.5 hour “long-haul” flight or four consecutive 0.5 to 1.5 hour “short haul” flights. Subjective fatigue ratings accurately reflected the rested and fatigued conditions; however, performance on a psychomotor vigilance task did not vary significantly between rested and fatigued conditions or between short and long haul flights. Further, although physiological data allowed a general distinction between larger and smaller fatigue effects for most individuals, no single measurement device was found to reliably indicate fatigue levels with enough granularity to allow for useful fatigue detection. However, a statistical/machine learning model was constructed that was able to accurately categorize fatigued data for each individual pilots using that pilot's combined sensor data with a success rate greater than 95%. It is probable that by optimizing the parameters of the model and improving the quality of sensor data, an algorithm-based solution to real-time fatigue detection can be developed. The applicability of this type of approach extends beyond the commercial flight deck to any work environment that requires multi-shift or other non-traditional scheduling.