Modeling Fatigue Associated with Workload in Commercial Aviation Operations

Introduction: Biomathematical models are widely used in commercial aviation to assess fatigue risk during operations. Most models are fundamentally influenced by the two-process model of sleep regulation, which estimates fatigue in relation to prior sleep duration and circadian rhythm/time of day. However, fatigue can also be due to workload. Many factors that contribute to workload are predictably related to scheduling and can be incorporated into a prospective model of fatigue. The biomathematical modeling software SAFTE-FAST recently introduced a Workload Calculator to allow users to estimate job demands during operations. Workload in SAFTE-FAST is modeled as a separate, but intervening, process from cognitive alertness related to time of day or sleep history. An important step toward accurate modeling of workload is to test the accuracy of the Workload Calculator against a real-world measure, like the NASA Task Load Index (TLX) scale. This study compared the accuracy of SAFTE-FAST Workload predictions against TLX scores taken by pilots during normal flight operations using equivalence testing.

Methods: Pilot participants from a major Asia-based airline completed the NASA TLX at top of descent (TOD) during the last flight of a multiple flight duty day on Day 1 and Day 3 of a three-day roster. Rosters consisted of daytime flights only to limit the influence of circadian fatigue or sleep debt on perception of workload. Pilots’ schedules were modeled in SAFTE-FAST. SAFTE-FAST Workload predictions and TLX scores were independently normalized to a 100-point scale. Workload changes across the roster were evaluated using repeated measures analysis of variance (ANOVA). Equivalence testing compared Workload against TLX using the two-one sided t-test (TOST) approach.

Results: Ninety-nine (N = 99) pilots completed at least one TLX over the course of the three-day roster; 88 pilots completed both TLX surveys for a total of 187 surveys. SAFTE-FAST workload estimates increased significantly across the roster (F(2,16)=62.02, p < 0.001). The average SAFTE-FAST workload prediction score at TOD for all flights was 64 ± 7 out of a possible 100 points (indicating high workload). The average TLX score at TOD for all flights was 65 ± 15 scaled to 100 points. There were no significant differences between Workload and TLX mean scores controlling for flight number or roster day. TOST results indicated that the means of Workload predictions at TOD were equivalent to TLX scores for the same TOD (t=1.56, p = 0.06).

Conclusion: Predicting the impact of workload on fatigue is a new frontier for biomathematical modeling. Establishing the accuracy of workload predictions is an important first step toward risk management in situations where high workload may create a safety risk. SAFTE-FAST predictions were statistically non-different from pilot-reported workload in this study, indicating model accuracy for predicting workload during operations. This is an important step toward modeling the impact of psychological factors (workload) on operator fatigue.

Publication History

Article published online:
08 October 2025