"It's my understanding HUMS picked up on the EC225 bevel failure in advance but due to the once per day download pattern used by the operator at that time, it went out on its afternoon flight and subsequently failed resulting in the ditch. Had the HUMS card been downloaded after its morning flight and before its afternoon flight, the impending failure was clear and would have been caught, the aircraft would never have departed. That's why the industry changed to more frequent after flight HUMS downloads in the aftermath.
If it's a gearbox failure this time around which to be honest, is more probable than probably not, I would expect the HUMS to come under close scrutiny. The difficulty the operators have is deciphering the data and what it actually means to the airworthiness of the helicopter, there are a lot of instrumentation defects for example so these need to be filtered out and it can be difficult to detect genuine mechanical failure modes. HUMS probably catches a lot that we are all unaware of because they get to it in time, it's only when failures happen, HUMS comes under scrutiny. "
It seems to me that for HUMS to truly be effective it needs built in learning software and immediate, internal analysis. i.e. Install it with a manufactured baseline data (expectation of vibrations for a particular aircraft type). Perform a few flights in the learn mode (gathering data about that specific airframe) review the data for those flights to make sure it isn't "learning" any vibrations that are of concern.
From that point on if it detects a vibration outside the standard deviation it could illuminate a light (similar to a chip light). On more complex systems it could even indicate where the vibration is on the display.
I believe all of the technology exists. How far are we away from something like this?