Good article there, however one point I would add is that the HUMS is like any other tool, it is only as good as the operator. My issue with HUMS has always been that there are several different flavours, dependant on aircraft type, and ultimately the HUMS data is analysed in detail rarely.. i.e. after an incident, otherwise the crews download the memory card, wait for the all clear from the groundstation and off you go. Whilst the "HUMS Guru's" at the likes of Bristow, CHC, Norsk etc are all very well trained and very capable guys, Their expertise can only be brought to bear on isloated incidents, simply due to workload, so it os entirely possible for a developing defect to be missed until it is too late, since HUMS sensors and more importantly vibration thresholds are programmed to look for a specific defect, within a specified frequency band.. anything outside of that, and its up to the HUMS Manager to spot using experience and a bit of system knowledge. I mentioned earlier the Dauphin incident, where a main ring gear in the MRGB cracked, this was a previously unheard of failure mode, but when the data was examined retrospectively, there it was.. for all to see. The upshot was that the monitoring system was updated to include the new failure mode, but only after a serious incident. (My use of "incident" rather than "Accident" is because I can't recall which aircraft was invloved.. French operator I think, but don't want to over dramatise)
Can anyone shed any light on the "Unsupervised Machine learning" HUMS technology that Smiths were working on? This was a system which used Neural networks and data cluster / mining techniques to constantly update its own thresholds. Did it ever evolve into a useable system? I ask because I have been out of North Sea ops for 3 or 4 years now, and I'm not up to speed with the very latest.
My point being monitoring is a good diagnostic tool, and it has contributed greatly to safety in offshore Helicopters.. Its just not fail-safe.