Probabilities
OK Woj I checked out this guy....
"His related research has resulted in patented early-warning systems for the jet engines of the Eurofighter, Airbus A380, and Boeing 787 Dreamliner, and for manufacturing processes that build such engines. He provides consultancy in machine learning to Rolls-Royce PLC and Oxford BioSignals Ltd., a university spin-out company."
He is employed as a full time "line" lecturer and teaches under grad basic math and post grad Bayesian math. I see he also has a strong interest in the medical field so I would guess this failure annoyed him from a moral viewpoint. A vocational person, for sure.
He also works as a consultant to RR.
In his paper
"Bayesian Extreme Value Statistics for Novelty Detection" in your link you posted the clue to this failure is very clear:
"An engine-specific approach to novelty detection has been
shown to be possible, in which the characteristics of data
from individual engines is learned – this often provides
better performance than generic “fleet-wide” models of
normality [21,22], in correctly identifying behaviour that is
abnormal for that engine, while minimising the number of
false-positive alarms generated."
In other words engines inherently are DIFFERENT. They are unique and they are seeking the normative values. Unfortunately there are not a lot of T900s around. T972s even less. They did not have enough DATA SETS. Clue number 1.
Now it is easy to quote sections of a paper to make a biased claim and I urge all readers to read the whole paper to make their judgement of just WTF is going on here, excuse me.[/LEFT]
"Future research will include the application of the technique
to vibration phase, and to non-vibration data sets, such as
“performance parameters” (engine temperatures, pressures,
etc.), which have previously been used with off-line density estimation
approaches [23]".
Well that about wraps it up. RR were running with this without the full data. The "novelty" threshold was set too high to minimise false alarms.
Basically they were taking a big risk.
I make it clear that the responsibility DOES NOT not rest with Mr. David A Clifton.
We will just see how long this post remains on this site. Some of us want the full picture.