BRENNEN

Industries / Transport and vehicles / Aviation / Helicopter services

Monitoring aircraft health

Monitoring aircraft health means measuring vibration, oil debris and usage on every flight so that a cracking gear, shaft or bearing shows up before it fails; the systems that do it are called HUMS, for health and usage monitoring systems.

7

Milestones achieved, from 1984 to 2024.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1984

    A panel asks for warning

    In June 1984 the UK's Helicopter Airworthiness Review Panel, formed by the Airworthiness Requirements Board, published 15 recommendations on helicopter certification standards, with emphasis on crashworthiness, survival at sea and the reliability of rotor and transmission systems, a transport minister told Parliament that November. After a Boeing Vertol Chinook crashed off Sumburgh, Shetland, on 6 November 1986, UK research money went chiefly into transmission monitoring, the Civil Aviation Authority wrote later.

    api.parliament.ukgov.ukcaa.co.uk

    Credit

    Helicopter Airworthiness Review Panel; UK Civil Aviation Authority

  2. 1999

    Vibration checks become compulsory

    Vibration health monitoring (VHM) puts accelerometers on gearboxes and shafts and turns their signals into condition indicators, numbers that each track the state of one part. UK and Norwegian operators fitted it to almost all large North Sea helicopters in the early 1990s of their own accord. The UK CAA made it compulsory from 1 June 1999 for UK passenger helicopters with more than nine seats; Norway's CAA required it for helicopters serving its petroleum industry from 1 July 2005, and EU offshore rules (SPA.HOFO.155) took over from 1 January 2019. The CAA says the accident rate from rotor, drive-system and engine failures in this class of helicopter has fallen significantly since, and that it believes VHM played a major role.

    caa.co.ukfaa.gov

    Credit

    UK CAA; Norwegian CAA; North Sea operators

  3. 2008

    Rotors lag behind gearboxes

    After a main rotor blade failed on a Sikorsky S-76A near the Leman 49/26 platform in the southern North Sea on 16 July 2002, killing 11, the CAA, Oil & Gas UK and Shell Aircraft paid GE Aviation to review rotor health monitoring. Its report, CAA Paper 2008/05, found that gearbox monitoring had matured but rotor monitoring had not, although the two kinds of failure cause comparable numbers of accidents, and that sensors on the fuselage were unlikely to catch some rotor faults early enough to help. It also warned that adjusting rotors for smoothness, called rotor track and balance, can mask a developing fault instead of revealing it. The CAA narrowed follow-on work to tail rotors.

    Lessona monitoring system watches only the parts it was built around.

    caa.co.uk

    Credit

    UK CAA, Oil & Gas UK and Shell Aircraft (funding); GE Aviation (study)

  4. 2010

    Learning what normal looks like

    From 2004 the CAA funded Smiths Aerospace, which became part of GE Aviation, to try "advanced anomaly detection" on HUMS data, working with Bristow Helicopters. Instead of a fixed alarm level for each indicator, the software learns what a fleet of healthy gearboxes looks like and flags data that does not fit, a form of unsupervised machine learning. GE's 2010 final report said two in-service trials showed it beat the usual threshold method at finding both helicopter faults and faulty HUMS instruments. When the AAIB had GE run it on G-REDL's data after that 2009 accident, it found no anomaly linked to the failed bearing.

    Lessonanomaly detection is only as good as the indicators it is given, the AAIB noted.

    assets.publishing.service.gov.ukdoi.org

    Credit

    UK CAA; Smiths Aerospace, later GE Aviation (Rob Callan, Brian Larder, Sue Salter); Bristow Helicopters

  5. 2013

    A gearbox check moves aboard

    On 10 May and 22 October 2012 two EC225 LP Super Pumas, G-REDW off Aberdeen and G-CHCN off Shetland, ditched in the North Sea after the shaft that drives the main gearbox oil pumps cracked; everyone escaped into liferafts and no one was seriously hurt. On both, MOD45, the HUMS indicator for that shaft, had crossed its learned amber threshold 4.62 and 4.75 flying hours before the oil pressure failed, but the data waited on a memory card for a download on the ground; G-REDW's alert the night before had been put down to a worn accelerometer washer. Eurocopter's service bulletin of 8 July 2013, made mandatory by EASA, had the aircraft compute MOD45 in flight and light a HUMS caption on the instrument panel when it crossed its threshold.

    Lessona warning stored for later download cannot help the crew on the flight in which a crack grows.

    assets.publishing.service.gov.ukad.easa.europa.eu

    Credit

    AAIB (investigation); Eurocopter, now Airbus Helicopters; EASA

  6. 2016

    Warnings missed on screen

    On 27 December 2016 the HUMS on a CHC Scotia Sikorsky S-92A, G-WNSR, recorded a series of vibration exceedances on the bearing of the tail rotor pitch change shaft. The engineer analyzing the download that night saw an anomaly on a related tail gearbox reading but not the exceedances, partly because of how the ground station software displayed them, and the aircraft was released. The next day yaw control failed as it landed on the West Franklin wellhead platform; it landed heavily and turned about 180 degrees on the helideck, and no one was hurt. The AAIB found that HUMS had detected the failing bearing and that the ground station's design was a contributory factor; Sikorsky brought in HUMS software with better diagnostics and screens.

    Lessona detection counts only when a person sees it in time.

    assets.publishing.service.gov.uk

    Credit

    AAIB; Sikorsky

  7. 2024

    HUMS written into maintenance rules

    On 9 December 2024 EASA amended CS-29, the certification standard for large helicopters (Amendment 12, ED Decision 2024/009/R), with new guidance for certifying vibration health monitoring so that it can be an approved part of maintaining rotors and drive systems. The rulemaking group, EASA and all CS-29 manufacturers, declined to require that the data be analyzed in near real time or shown to the crew before takeoff and after landing, as the AAIB had recommended after G-WNSR; the AAIB rated that response "Not Adequate" in January 2026. EASA also closed, in December 2025, the AAIB's call for research into capturing data more often, saying current systems complete a cycle within minutes in steady flight.

    assets.publishing.service.gov.uk

    Credit

    EASA; AAIB

Watch for

  1. Anomaly scores from healthy data

    Researchers at Politecnico di Milano and TU Eindhoven, with three authors from Leonardo Labs, part of the helicopter maker Leonardo, describe a method that learns only what a healthy transmission's vibration looks like, gives a probability that new data is abnormal, and offers tools to explain each alarm. They report "competitive detection performance" on a public benchmark and on a real helicopter transmission dataset collected over several years, which is not public.

    Achieved whenan independent team reproduces the result on public gearbox data, such as the Australian Defence Science and Technology Group's planet gear datasets, or an operator or regulator reports the method in service with its false-alarm rate.

    arxiv.org

    Claimed by

    Politecnico di Milano and Leonardo Labs, March 2026 (arXiv preprint, not peer reviewed)

Open

  1. 01

    HUMS warnings timed from reports

    A public table of every investigated rotor or drive-system failure on large helicopters since 1997 whose final report discusses monitoring data. Each row gives the aircraft, date, part, failure mode, whether HUMS was fitted, which indicator moved, the flying hours from the first threshold crossing to the failure, how often data was downloaded, whether anyone saw the alert, and a link with page numbers. Add a replay that shows, case by case, whether checking data after every flight, every three flight hours, or on board in real time would have caught it.

    The brief
    Credit · unclaimed
  2. 02

    Test HUMS screens with people

    An open-source mock-up of a HUMS ground station that shows a day's indicator trends for a small fleet, fed with synthetic data containing planted exceedances, and a timed test in which engineers or engineering students review downloads on a screen like today's and on redesigned ones, one of which adds an AI-written summary of each alert.

    The brief
    Credit · unclaimed

Open 01 · the brief

HUMS warnings timed from reports

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of every investigated rotor or drive-system failure on large helicopters since 1997 whose final report discusses monitoring data. Each row gives the aircraft, date, part, failure mode, whether HUMS was fitted, which indicator moved, the flying hours from the first threshold crossing to the failure, how often data was downloaded, whether anyone saw the alert, and a link with page numbers. Add a replay that shows, case by case, whether checking data after every flight, every three flight hours, or on board in real time would have caught it.

Why it is still open

The lead times exist but sit in separate reports: 4.62 and 4.75 flying hours for the two 2012 EC225 shaft cracks, about a day for G-WNSR, none for LN-OJF or G-REDL. The AAIB and EASA still disagree over whether crews should see exceedances before takeoff, and EASA closed the question in 2025 without requiring it. The CAA's 2008 review gave case histories for eight rotor failures up to 2006 but not for drive systems or later events. Cranfield University's study for EASA (research report 2012/6) ran fault-tree analyses on 12 accident reports chosen from 413 and found "no consistent patterns", according to the safety consultancy Aerossurance's summary, which gives no lead times; no public cross-report table was found.

assets.publishing.service.gov.ukcaa.co.ukassets.publishing.service.gov.ukaerossurance.com

Built with what exists

Final reports from the AAIB, Norway's NSIA (formerly AIBN), TSB Canada, the ATSB, the NTSB, the BEA and CENIPA; pdftotext or PyMuPDF; an AI model such as Claude Fable 5.2 (see the Note "Three working devices") filling a fixed form from each report, with every value checked against the page by a person; WebPlotDigitizer to read indicator plots; Python for the replay.

  1. First step

    Code the five North Sea reports already public in English (G-REDL, G-REDW, G-CHCN, G-WNSR, LN-OJF) and fix the field list.

  2. Done looks like

    At least 30 events from at least five investigation bodies, each coded on at least ten fields with page references, 20% coded again independently by a second person with the agreement reported, and the replay and its code published under an open license.

Open 02 · the brief

Test HUMS screens with people

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open-source mock-up of a HUMS ground station that shows a day's indicator trends for a small fleet, fed with synthetic data containing planted exceedances, and a timed test in which engineers or engineering students review downloads on a screen like today's and on redesigned ones, one of which adds an AI-written summary of each alert.

Why it is still open

The AAIB found that G-WNSR's exceedances were missed partly because of how the ground station displayed them, and named the screen design a contributory factor. Sikorsky has since changed its software, but a Crossref search found no published head-to-head test of HUMS screen designs; the nearest was a 2002 paper on linking a ground station to a maintenance records system.

assets.publishing.service.gov.ukdoi.org

Built with what exists

Python with Plotly Dash for the screens; synthetic indicator histories whose crack trends are shaped after the Defence Science and Technology Group's planet gear test (its own data may not be redistributed); Claude Fable 5.2 (see the Note "Three working devices") for the alert summaries; a simple web timer and log.

  1. First step

    Rebuild the screen described in the G-WNSR report's ground station section and run it with five volunteers to confirm the task takes under an hour.

  2. Done looks like

    At least 24 participants each review 40 aircraft-days containing 8 planted exceedances on two or three screen designs in random order, with hit rate, false calls and time per download published together with the code and data.