Software Bug Causes Soyuz To Land Way Off

howhardcanitbetocrea writes "A mysterious software fault in the new guidance computer of the Soyuz TMA-1 spacecraft was the cause of the high-anxiety off-course landing over the weekend, according to NASA sources.' Which is why I will never trust the Strategic Defence Initiative - the star wars project. It only takes one line of mistyped code in what will always be a beta release."

Comments

Fail-safe design

[fname]

It's actually a clever piece of work. Basically, software has to make calculations in order to provide a "soft" entry, 5 Gs approximately. If there is an error, the module goes into a ballistic entry mode, and it is more like 7-8 Gs, rougher but survivable.

On (nearly) every manned spacecraft ever flown, every system has a hot-backup that kicks in if the first one fails. The exceptions are systems for which it is basically impractical to have a backup-- can't really have redundant heat shields, as the weight is too much. But for electronics and software, this is standard. This story would have gone practically unnoticed if Soyuz had notified Star City that they were doing a "ballistic" entry, in which case they would have been located much sooner.

This landing showed that the Soyuz has a robust design; if Endeavour enters the atmosphere at the wrong angle, could it recover? What if the flight landing computer failed? NASA has a lot of these things covered; for many problems it is probably more robust than Soyuz, for others it is less robust. Soyuz has the advantage of much more flight experience; I doubt that it's a coincidence that this anomaly happened on a flight with a newly upgraded Soyuz.

Yeah, but still no integration test

[enkidu]

What you're talking about is component level testing. Unfortunately, all that testing doesn't substitute for a true "shakedown" integration test. Look up the AEGIS cruiser system (actually sort of a mini-SDI for a ship). On it's first full integration test, it failed to shoot down 6 out of 17 targets due to software errors. Now, make the integrated platform 2 orders of magnitude more complicated than that (and at least one order of magnitude more complicated than ANY software project attempted to date) and you can see why I'm skeptical of the chances of SDI working as advertised.