A fatigue risk management system that ties every fatigue report to its roster and the science the same day
Every fatigue report tied to its roster and the science the same day, fixed by the board.
See one case, screen by screen ↓A fatigue risk management system (FRMS) is the data-driven way an airline finds and fixes the causes of crew fatigue, as ICAO Doc 9966 and, in the US, 14 CFR 117.7 describe it: collecting fatigue reports, linking them to the duties that caused them, assessing the risk, changing schedules and proving the change worked. The Fatigue Risk Management System is FRMS software that puts each crew fatigue report, the roster actually flown and a biomathematical fatigue model on one time axis, with eight agents that link, model, find patterns, draft cases and test legal pairing fixes, while the fatigue risk manager and the FRMS review board decide.
How to choose AI for airlines: a buyer's guide to agentic AI solutions →
Updated
Part 117 says a pilot too fatigued to fly may not fly, and every Part 121 airline must keep a Fatigue Risk Management Plan, train crews on fatigue every 12 months and update the plan every 24 months. Flying beyond the Part 117 limits — an ultra-long-haul route, say — needs an FAA-approved FRMS under 14 CFR 117.7. At the heart of both is the same job: a crew member says “I was too tired”, and the fatigue safety team has to find out why and fix the cause.
Today the fatigue report sits in a safety form, the roster in crew scheduling, the actual out, off, on and in times in operations, and the fatigue model in a separate modeling tool that scores the planned roster, not the one flown — with an assumed layover sleep, not the sleep the crew member wrote down. Linking them takes an afternoon per report, so most reports are acknowledged, filed and counted. Eleven reports on one red-eye show up as eleven separate lines, and the pattern surfaces months later at the quarterly review.
Estimated split of a fatigue safety team's working hours for one case, from the first report to a board decision pack, by hand and with the solution.
Eight agents read, link, model, cluster and draft every fatigue case; the fatigue risk manager decides and the review board signs.
Corvane Airways' red-eye CV 2290 LAS–DEN, day two of pairing D4417 on the A320neo out of the Denver base. On Sunday 18 October a first officer filed fatigue report R-26-1187: how tired 6 of 7 at top of descent, about 3 h 40 of broken sleep, a missed frequency change on the arrival. It is Tuesday 20 October, 07:40 Mountain time. Here is what happens next, screen by screen, in the working solution.
Megan's Today screen opens on four numbers: 12 new reports, 23 open cases, the review board on Thu 22 Oct and the FRMP update due 3 Nov. The lead item is the one that needs a decision before Thursday: red-eye CV 2290 LAS–DEN, 11 fatigue reports in 60 days on one pairing, a case already drafted and four pairing changes tested. Beside it, the agents brief her as a conversation, each line cited.
Report reader: “Four need a person: an illness that may not be fatigue, a duplicate, a deadhead that is not in the roster, and one also filed as a safety report.”
The FRMS case opens on Sat 17 – Sun 18 Oct in three lanes that share one axis. In her words: “could not sleep in the afternoon”, “fighting to keep my eyes open”, pinned at the times they describe. The roster actually flown: CV 1187, a rest of 11 h 00 — legal — and CV 2290 inside the body-clock low. And alertness, two curves: the model with the sleep it assumes, and the model with the sleep she reported.
Fatigue model: “With the layover sleep the model assumes (6 h 10), alertness at landing is 66. With the 3 h 40 she reported, 58. The duty is inside the model’s validated range.”
The report as filed sits beside what the Report reader took from it, each field with its confidence and source: CV 2290 on Sun 18 Oct, pairing D4417 day 2, how tired 6 of 7, sleep 3 h 40 broken, an afternoon layover in a pool-deck room, the third time this month. The reporter appears only as R-7731. Only the gatekeeper can reveal a name, with a reason code, and the reveal is logged and reported to the board chair and the fatigue committee.
“Only the gatekeeper (Megan Ashford) can reveal it, with a reason code. The board sees codes only.”
The Pattern finder groups every report by pairing, flight, hotel, report time and aircraft, and raised this cluster on Sun 18 Oct at 07:04: 11 reports in 60 days, 9 pilots and 2 flight attendants. Nine pilot reports in 120 pilot duties is 75 per 1,000 against 4.1 for the fleet. Asked what else could explain it, it rules out the same few crew (11 reporters), delays (CV 2290 was on time on 52 of 60 nights) and the aircraft (9 tails).
Pattern finder: “What the reports share is the afternoon release and the pool-deck rooms.”
The Case writer drafted the hazard — an afternoon layover in LAS before a red-eye that lands in the body-clock low — rated the risk 4C, high, and recommended option A. Then the Case checker re-opened all 23 sources behind the draft: 22 confirmed, and one claim softened because two reports give no sleep figure. Every step read only; nothing was changed on a roster or pairing.
Case checker: “‘crews report 3–5 h of sleep’ now reads ‘9 of 11 reports give 3 h 40 to 4 h 55’.”
The Pairing planner built four options for D4417 with the pairing optimizer and scored each on the crews' reported sleep. A: arrive LAS the evening before, a 28 h 40 layover — alertness at landing 58 → 74, $214 per operation, $78,110 a year. B: the red-eye as its own pairing — 72, $124,465 a year and 62 more pilot days a month. C: retime CV 2290 to 23:05 — 77, but it moves a revenue flight, so it goes to Network Planning as a note. D: a later DEN–LAS with a nap window — refused.
Option D: “117.25(e): 10 consecutive hours of rest before a flight duty period” — its rest would be 6 h 45.
Beside the similar-reports table, Megan asks the case agents: “Why option A over option B?” The Pairing planner answers with its sources: both lift alertness above 70, A to 74 and B to 72; A costs $214 per operation and B $341 with 62 more pilot days a month; C scores best but is not crew planning's call; D breaks the 10-hour rest rule.
Megan sends FC-26-0418 to the board for Thu 22 Oct, 10:00, with reason C2 — pattern confirmed, schedule change needed — and her electronic signature. The decision pack carries the 11 reports de-identified, the time axis, the model runs, options A–D and their cost. The Case writer adds it to the agenda as item 4 and tells the five voting members.
“The board decides; Crew Planning publishes only what the board approves.”
The board sees one page for the red-eye: 11 reports de-identified, alertness at the low 58 → 74, about $78,000 a year, the recommended option and links to the time axis and all four options. Members vote as themselves with a reason code; quorum is 3 of 5 voting members with at least one crew representative. When it is approved, Jordan Pike publishes option A for the 1 Dec 2026 bid period — and only he can.
Every change is measured for 90 days: reports per 1,000 duties, the alertness low on the duty and fatigue calls. On the assurance screen, six changes are working, one is too early to tell, one is still collecting data and one is not working: ending 737 four-leg days before 23:00 at LAX left reports at 5, so the Assurance monitor suggests re-opening the case. A case closes only when a person says the change worked.
“A sudden drop in reports can mean crews stopped reporting, so the monitor also tracks fatigue calls.”
The same red-eye, seen by five of the people who answer for it — what fatigue work looked like, and what it looks like now.
Eight agents read, link, model, cluster and draft every fatigue case; the fatigue risk manager decides and the review board signs.
Reads every fatigue report from the crew app, email and paper forms; extracts flight, duty, sleep, how tired and contributing factors; seals the reporter's identity and flags edge cases — illness, duplicates, no roster match, a cross-filed safety report.
Pulls the 7 days of roster actually flown before each report and the actual out, off, on and in times; computes duty, rest and body-clock-low exposure against the Part 117 fatigue rules and 121.467.
Runs the biomathematical fatigue model on the actual roster twice — with planned sleep and with the sleep the crew member reported — and explains the difference.
Clusters reports by pairing, flight, hotel, report time and aircraft, compares each cluster with the fleet rate and raises it when it reaches the rule — 5 reports in 60 days.
Drafts each FRMS case — hazard, evidence, risk rating on the FRMP matrix and mitigation options with their predicted effect — and builds the board's decision packs and minutes.
Re-opens every source behind a draft or a pack and refuses claims it cannot find, softening or marking anything it could not verify.
Builds pairing change options with the crew pairing optimizer and scores each for alertness, Part 117 legality, the labor contract and cost per year; options that break a rule are shown refused, with the rule.
Measures every change for 90 days, tracks the FRMP 24-month and fatigue training 12-month clocks, drafts the FRMP update and builds FAA evidence packs.
Decides; the review board approves schedule changes. The agents propose; a named person decides.
Ask about any report, case, pairing or clock — on the Today screen, inside a case or on a pairing. Every answer cites the report, roster, model run or rule it came from.
Why is CV 2290 a fatigue risk if the rest is legal?
The rest is 11 h 00, legal under 117.25(e), but it starts at 13:45 in Las Vegas, when the body does not want to sleep. The first officer slept 3 h 40. With that sleep, alertness at landing is 58, below the high-risk line of 62.
Which reports need me today?
Four: R-26-1192 may be illness, not fatigue; R-26-1199 duplicates R-26-1190; R-26-1193 deadheaded on another airline, which the roster does not show; R-26-1190 was also filed as a safety report, so I keep the two apart. I proposed a path for each; you decide.
Why option A over option B?
Both lift alertness at landing above 70: A to 74, B to 72. A costs $214 per operation, B $341, and B adds 62 pilot days a month. Option C scores best (77) but moves a revenue flight, so I sent it to Network Planning as a note. D breaks the 10-hour rest rule.
Who is due fatigue training in the next 30 days?
362 people: 119 pilots, 214 flight attendants, 14 dispatchers, 9 crew schedulers and 6 managers. 14 have lapsed and are off the roster until trained.
13 screens from the working solution, on its sample data. Pick one to see it large.
How the fatigue risk management system scores, raises and watches — set by the FRMS administrator, logged, and changed only with a reason.
| Setting | Default | Choose from |
|---|---|---|
| Elevated alertness below | 70 | 60 to 80 |
| High alertness below | 62 | 50 to 70; below it the case is rated high and goes to the board |
| Pattern: reports | 5 | 3 to 10 on one pairing, flight or hotel |
| Pattern: window | 60 days | 30 to 120 days, in steps of 15 |
| Also raise a pattern at | 3 reports | 2 to 6 reports rated 6–7 on one duty |
| Monitor every change for | 90 days | 60 to 180 days, in steps of 30 |
| Acknowledge a report | 24 hours | Thank the reporter through the channel |
| First draft · board decision | 72 hours · 35 days | From the pattern being raised · from the draft being sent |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published rules (14 CFR 117.3, 117.5, 117.7, 117.9 and 117.25; Public Law 111-216 §212 — FRMP updated every 24 months, fatigue training every 12; FAA InFO 15001 and AC 117-2 on fatigue education and awareness training; 14 CFR 121.467; FAA AC 120-103A; ICAO Doc 9966). Corvane Airways, Larkspur Air, and the people, reports, pairings and hotels named on this page are fictional — characters and sample data in the working solution.
A fatigue risk management system (FRMS) is the data-driven way an airline finds and fixes the causes of crew fatigue, as ICAO Doc 9966 and, in the US, 14 CFR 117.7 describe it: collecting fatigue reports, linking them to the duties that caused them, assessing the risk, changing schedules and proving the change worked. The Fatigue Risk Management System is FRMS software that puts each crew fatigue report, the roster actually flown and a biomathematical fatigue model on one time axis, with eight agents that link, model, find patterns, draft cases and test legal pairing fixes, while the fatigue risk manager and the FRMS review board decide.
Every Part 121 carrier must keep an FAA-accepted Fatigue Risk Management Plan (FRMP), update it every 24 months and train on fatigue every 12 months. An FAA-approved FRMS under 14 CFR 117.7 is what an airline needs to fly beyond a Part 117 limit, such as an ultra-long-haul route. The solution supports both: it drafts the FRMP update and tracks the training clock, and it keeps the validation data for a 117.7 approval — in the demo, JFK–SIN at month 7 of 12.
The Fatigue model agent runs the model on the roster actually flown twice: once with the sleep the model assumes and once with the sleep the crew member reported. In the demo, a legal 11-hour rest that began mid-afternoon gave 3 h 40 of sleep, so alertness at landing is 58 instead of 66. The agent says when a duty is outside the model's validated range, and a score is evidence, not a verdict.
Yes. Reporters appear as codes on every screen, decision pack and export. Only the gatekeeper — the fatigue risk manager — can reveal a reporter, with a reason code, and the reveal is logged and reported to the board chair and the fatigue committee. Follow-up questions go through the crew app's confidential channel.
The Pattern finder groups every report by pairing, flight, hotel, report time and aircraft, compares each cluster with the fleet rate and raises it when it reaches the rule — 5 reports in 60 days, or 3 rated 6–7 on one duty. It raises patterns but never opens a case by itself; a person drafts, watches or dismisses each one with a reason.
No. The Pairing planner builds and scores options for alertness, Part 117 legality, the labor contract and cost, and shows any option that breaks a rule as refused. Schedule changes need the FRMS review board's vote, and only the crew planner publishes the approved change for the next bid period.
Yes. The agents read, link, model, draft and propose; they never reveal an identity, send a case to the board, vote, publish a pairing or close a case. The fatigue risk manager signs each decision with a reason code, the board votes with a quorum of 3 of 5 including a crew representative, and every action is on a chained audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your FRMP, risk matrix, reason codes, contracts and a history of past reports — and it goes live once every quality gate has passed. In the demo, first-time setup connected crew scheduling and operations data and turned on confidential reporting in the crew app, set the fatigue model thresholds at 70 and 62 against 1,240 past reports, and re-read and linked two years of past reports — 4,488 — finding 17 patterns.
Every broken pairing re-crewed legally before it costs a flight, crew told in 20 minutes.
1-min demoFlight and Duty Time ComplianceNot one illegal duty day published, and every limit shown with the rule that sets it.
1-min demoCrew Pairing OptimizationEvery trip legal and cited, at the cost and crew quality of life you chose.
1-min demoWe'll run the Fatigue Risk Management System on a past quarter of your own fatigue reports, rosters and actual times.