Hacker News Reader

Comment by GekkePrutser | Hacker News Reader

That's crazy, in the Netherlands today there was another flight that dropped engine parts (much smaller though, they're said to have been turbine blades) that hit cars and people. It was also an airliner (747-400).

Considering how rare an occurrence is, it's really weird to have both happen on the exact same day.

https://www.airlive.net/breaking-turbine-blades-of-a-boeing-...

By the way I'm surprised that fire keeps going on the B772. I assume the fuel was cut off, especially as the turbine seems to be just spinning in the wind. I'd imagine the remaining fuel would burn off quickly?

This would really be scary being a passenger (especially when you know the wing it's hanging under is full of fuel).

View on HN

Random = clusters happen.

In fact, mathematicians were hired during the Battle of England to estimate whether the Germans were aiming for specific buildings (“why is there a cluster of bombs around this church???” or around some secret offices) or just spreading randomly. Verdict: They weren’t aiming, it was random chance that bombs made some clusters onto some buildings. This is what random looks like: No even distribution, sometimes you get a cluster which looks like a series.

I think it depends on how frequent the thing in question is. If an engine losing parts is something that happens yearly, twice in one day is still pretty rare.

But then you have to deal with degrees of freedom as well. If your search space is just 'two uncorrelated ~yearly events', there's enough of them that you'll always find false positives.

For independent, random occurrences, the probability distribution for delta-T follows an exponential decay. I.e. the most likely time is "right after". Of course, that "most likely" can still be small. Fun fact: This is true for any chosen point in time, i.e. the prob that this happens exactly one day after your birthday is higher than that it happens two days later.

I've tried to find a non-clustering pseudo-random function.

Anybody knows the name or urls of those?

Maybe quasi-random is what you are looking for?

https://www.gnu.org/software/gsl/doc/html/qrng.html

There’s the (descriptively named) “low-discrepancy sequence”, the most famous one being the Sobol sequence.

https://en.wikipedia.org/wiki/Low-discrepancy_sequence

This is exactly it! Many thanks for the name.

This is my go-to: http://extremelearning.com.au/unreasonable-effectiveness-of-...

Where I have read this before, Gravity's Rainbow?

Grinstead & Snell in the part on Poisson distributions maybe?

Poisson distribution, higher probability of short periods between failures than long ones.

Anyone care to elaborate? I don't understand what he means, but I feel like this is an interesting thread to pull.

Briefly: if you expect 10 failures in 10 years evenly distributed then you would expect 1 per year.

If it is a Possion distribution then you'd expect them to bunch up and get years with none, then them concentrated over a few years.

Possion distributions usually imply a time factor. That's concerning in this case as it implies maintenance isn't effective.

how would this apply to disparate aircraft on different maintenance cycles?

It has nothing to do with aircraft, and everything to do with random chance. Basically, it is more likely for time-series data to have clumps than to be perfectly even.

Imagine you throw a three darts a board. Which is more likely: all darts are equidistant, or 2 of the darts are closer to each other than the 3rd? Now imagine it's 3 events in time rather than 3 darts on a board. And the same logic applies to an arbitrary number of events, of course.

it seems to me that there is a lot of room between "perfectly even" and "twice on the same day"

https://en.wikipedia.org/wiki/Clustering_illusion

I read on the German news about the Dutch incident and came to HN because I knew that somebody would've posted about it here. Found this thread and didn't fully read the headline, just "Boeing" and "engine failure". It took me a few seconds to realize that this was a completely different incident.

The Dutch incident is a relatively common one, the reason it's in the news is that turbine blades fell down on a residential area and someone got minor burns from picking one up.

It's more typical to have this kind of failure early in the take off where the parts would fall on the airport and nobody would talk about it.

All the rest of the news about circling at 10,000ft and then landing at a nearby airport is completely standard procedure that gets practiced all the time in the 6 months simulator checks. Basically you climb high enough to be able to dump fuel without it reaching the ground (minimum 6000ft, preferably higher) and still not too high such that you're burning a lot of fuel quickly because at 10,000 ft fuel usage is way higher than at 30,000. Then when the fuel dump is complete, engine secured, checklists done it's a relatively normal landing. Especially if you have 4 engines but even with 2 it's a landing every pilot can do and is trained for all the time.

It could be oil that's on fire

There is a saying, something like "once is bad luck, twice is a coincidence, three times is enemy action". So, if you get more than two big planes crashing in the same day, it's likely something like Al Qaeda (sp?) on 2001/09/11. If it's two planes, it's likely bad luck.

Reply on news.ycombinator.com