Schiaparelli probe crash caused by excessive spin, report concludes The Schiaparelli probe suffered a botched landing on the Martian surface as it briefly spun out of control, confusing the computer systems onboard, an official report concluded today. As the parachute was deployed, the lander rotated at a faster acceleration than expected. It caused the “Guidance Navigation and Control” …
Drop a dummy load with the same shape/size/weight/balance from a plane over the desert, with any thrusters or parachutes adjusted/adapted to Earth's gravity and atmospheric density. Sure, it isn't an apples to apples test when compared to actually landing on Mars, but it would have caught a lot of the errors (like the infamous metric versus English unit conversion) that have tripped up previous Mars missions, and would stand a good chance of catching Schiaparelli's problem as well.
The problem is if the parachute is the thing you're trying to test, altering it for Earth's much thicker atmosphere might invalidate the test.
Yep. Curiosity's record-setting supersonic parachute had numerous failures on Earth. It kept breaking in wind tunnels because the conditions differed from Mars. Curiosity's parachute obviously worked on Mars. Mars Rover 2020 also had some live testing failures in 2015.
You could make a parachute that would be a pretty close approximation in terms of behaviour to that of the other one on Mars. And as it would be noticeably smaller and could be tested by firing it from a gun, lots of times for not a lot of money.
You could, but then you're not testing the same item you're using in production. And scale matters a lot in aerodynamics, because you can't scale down the air molecules. This is especially important with things designed to create drag, like parachutes, because they create turbulent flow and the scale that turbulence works on is pretty fixed.
They tested as much as was possible on Earth, but without an unlimited budget, at some point they had to stop testing and try a trial run, actually on (or rather, above) Mars.
Something almost as important as testing, is proofreading what you've just written. Just saying...
However, choosing those limits is probably quite difficult.
Rejecting negative altitudes wouldn't seem that difficult, nor would "enveloping" each parameter to exclude obvious outliers. The real problem is whether there's sufficient different data to inform a choice of which is most probably correct, and what to do next when the systems receive data that cannot be trusted.
Scientific exploration is always risky, but the loss of Schiaparelli still seems to have a ring of undue carelessness in a €230m project. Conflicting or palpably wrong data has been a control issue since men first started fitting sensors on machines. And it isn't as if there's plenty of precedents, particularly in aviation - the root cause of the loss of AF447 in 2009 was the inability of the automatic control systems to cope with sensor data conflicts. And modelling control systems on Earth should be amongst the easiest parts of the whole task - but as we all know, software testing is the dullest, least rewarding, most under-appreciated part of any code-related exercise.
lander rotated at a faster acceleration than expected
Some similarities to the Ariane 5 Explosion, where an accelerometer value in a 64 bit floating point value overflowed when converted to a 16 bit signed integer.
Whenever I read of one of this hilarious snafus where someone manages to put a stupid number of email addresses in the "To:" field, I find myself wondering why there aren't mechanisms
1) to alert the user if more than <x> addresses are in the "To" field
2) to configure the server to prevent - or at least flag up - instances where an email is sent to more than <x> people ...
similarly, there should have been something in the code to reject out of bounds readings.
Now when I were a lad (30 years ago) we learned to buffer inputs from devices, and compare them over a period - maybe average out the last 10 readings. Why ? To avoid a spike triggering an action. I have to ask if there was something like this being used or not ?
"Very close to a success" or as human beings say "fail". Why is is so difficult for Schiap Chaps to admit this? Someone fucked up.
And as usual, that someone was a programmer, sorry, software architect, who in a move of staggering incompetence wrote a critical routine that had no "gibberish result" check.
Nought out of ten. Go and read "Object Oriented Software Construction" and fucking well pay attention to the basic rule about unspecified results.
"Nought out of ten. Go and read "Object Oriented Software Construction" and fucking well pay attention to the basic rule about unspecified results."
Bah x2! When was the last time you pulled off a project this complex with zero screw ups that required patching? This is an incredibly complex project with what by normal standards a deficient test capability.
Ego stroking won't improve anything. Any dickhead can do that. Running a witch hunts is known to increases future failures. You do the full and frank analysis, incorporate changes to systems and process and move forward. To do that, you need the right environment, ie. you need to put a bit of ego aside.
Complicated???
What the fuck is complicated about defining the legal domain of an altitude calculation to be a POSITIVE INTEGER?
Or in using a violation of that to start the process of taking measures to, you know, not plough into the Martian regolith at maximum wellie?
We've known how to tell negative numbers from positive ones in code for a few years now, way before the commissioning of The Martian High-Cost, High-Speed Impacter.
My original outraged comment stands.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
