World Sleep Industry ENCYCLOPEDIA

Pilot Fatigue Risk Assessment Systems

4.2.4

System purpose, operational scope, and regulatory boundary

A pilot fatigue risk assessment system is a decision-support tool that turns rosters, flight and duty activity, sleep opportunity, circadian timing, operational change, and personnel reports into reviewable risk judgments. It should sit within an operator's safety management system or a fatigue risk management system accepted by the relevant authority, not operate as a standalone score. Scheduled passenger, cargo, charter, aeromedical, helicopter, general aviation, and single-pilot operations differ in crew complement, predictability, workload, and applicable rules. The system must identify the operations it covers and the basis on which its use is approved. Prescriptive flight and duty limits are a baseline, but compliance does not establish restorative sleep; conversely, a high model score is not automatically a violation or an aeromedical finding of unfitness. The system cannot approve a deviation or duty extension by itself or replace the lawful judgment of the pilot in command, dispatch, operational control, aeromedical professionals, or the regulator.

The complete operational timeline from roster to release

Assessment should start with sleep opportunity and commuting before report and include reserve or standby, preflight preparation, every sector, ground waits, delays, diversions, returns, training, administrative work, positioning, and final release, followed by recovery. Data must distinguish planned and actual time, local and common time, home-base time zone, midnight and daylight-saving transitions. Relevant features include sectors and takeoff-and-landing workload, consecutive night operations, time-zone displacement, acclimatization state, early starts, late finishes, and cumulative duty. Standby notification and facilities, transport to away-base lodging, authorized rest in a cabin or flight deck, augmented-crew rotation, and the location of positioning all alter real sleep opportunity. Nominal rest consumed by commuting, calls, hotel delays, or other work cannot be counted wholly as sleep. Each reassignment and crew change should trigger recalculation, with the source of inputs and reason for modification retained.

Biomathematical models, assumptions, and limits of use

Biomathematical models commonly use sleep-wake history, circadian phase, and time awake to predict population-average alertness or performance risk. They can compare roster options, locate higher-risk windows, and explore mitigations. They do not directly measure the present fatigue of an individual pilot and may incompletely represent workload, illness, stress, sleep quality, individual differences, cabin conditions, and operational complexity. A system should expose the model and version, target population, input window, missing-data treatment, sleep assumptions, time-zone and acclimatization logic, output unit, and uncertainty. A color or threshold must not be presented as a precise personal safety boundary. Model selection needs aviation-relevant data and independent evidence, while calibration must not be optimized merely to make rosters pass. Outputs have to be interpreted alongside evidence-based scheduling, experience from comparable operations, pilot reports, observed sleep, and safety data. They must never be the sole basis for release, rejection of a report, or discipline.

Predictive, proactive, and post-event assessment

A mature system maintains three evidence paths. Predictive assessment examines rosters before publication for night work, consecutive early starts, rapid time-zone transitions, short recovery, and multi-sector combinations. Proactive assessment compares planned with actual duty in live operations, receives fatigue reports, surveys, or sleep diaries, and observes delay and workload. Post-event assessment asks whether fatigue contributed to a deviation, near miss, or accident. A pilot who believes fatigue may affect safety before report or during an operation needs a clear, usable, confidential path to report and stand down without inappropriate retaliation; a low system score cannot overrule that report. Reporting should collect the minimum information relevant to a decision and connect promptly to relief, reassignment, delay, cancellation, or another approved mitigation. Commander discretion or an emergency exception applies only in the special circumstances defined by the governing rules. The reason, crew views, actual extension, and subsequent recovery need records, and an algorithm must not recommend discretion as routine cover for understaffing.

Roster mitigation, in-flight rest, and fitness support

The preferred control changes the source of risk: adjust report time, reduce high-risk sequences, add recovery, improve sector combinations and standby notice, use an augmented crew, or provide suitable lodging and transport. The benefit of in-flight rest depends on the facility class, timing, rotation, noise, turbulence, and whether sleep occurs. The system should distinguish an allocated rest opportunity from sleep obtained. Controlled rest on the flight deck may be used only where applicable rules and the operations manual authorize it, workload permits it, and a complete procedure is followed. It is not statutory rest and must not be used prospectively to justify extending duty. Persistent sleepiness, possible sleep-disorder signs, difficulty adapting to shifts, or health and medication effects should lead to a confidential and accessible peer-support, occupational-health, or aeromedical pathway. An algorithm cannot diagnose disease or automatically decide grounding and return to flight. Fitness is a shared responsibility of the organization and the pilot, not a means of transferring the consequences of unsafe scheduling to the individual.

Safety assurance, metrics, and management of change

Before deployment, the system should be validated with representative operations: east-west and north-south travel, night cargo, multi-sector short haul, augmented long haul, reserve activation, severe delay, base change, and weak connectivity. Safety assurance should combine fatigue reports, planned-versus-actual variance, sleep or survey data, operational events, flight-data-monitoring signals, and audit findings to determine whether controls reduce risk. Metrics need exposure denominators and reporting context. More reports can indicate greater trust, while a lower average score can conceal a small set of hazardous duties. Investigations should preserve the model, parameters, roster version, actual timeline, mitigation decision, and crew views that existed at the time and examine organizational and operational contributors. A new route, base or aircraft, a standby-policy change, system integration, model revision, or rule update requires prior impact assessment. If safety performance is not maintained, the operator must correct controls, review thresholds, and where necessary suspend the use and roll back safely.

Data governance, procurement, oversight, and exit

The system may combine pilot identity, roster, location, sleep, self-assessment, health clues, reports, and performance data. Purpose limitation, data minimization, tiered permissions, retention periods, and traceable correction are required. Rostering, operational control, safety, aeromedical services, human resources, regulators, and vendors should not automatically see the same detail. Data used for safety learning should be de-identified and aggregated where possible, and individuals need routes to access and challenge it. Procurement should verify rule configuration, model evidence, time-zone algorithms, data lineage, interface failure, offline operation, cybersecurity, signed versions, change notice, and human review. Contracts should also permit regulatory oversight and export of investigation evidence. A vendor cannot invoke trade secrecy to prevent the operator from understanding material assumptions and limitations. On replacement or exit, the operator should migrate statutory records, model versions, open risks, and audit trails, securely delete data without a continuing basis for retention, and preserve the ability to reconstruct historical decisions.