Ariane 5 Flight 501 Robustness Failure _
June, 1996 loss of inaugural flight
- Lost $400 million scientific payload (the rocket was extra)
Efforts to reduce system costs led to the failure
- Re-use of Inertial Reference System software from Ariane 4
- Improperly handled exception caused by variable overflow during new flight profile (that wasn’t simulated because of cost/schedule)
- 64-bit float converted to 16-bit int assumed not to overflow
- Exception caused dual hardware shutdown (because it wasassumed software doesn’t fail)
What really happened here?
- The narrow view: it was a software bug -- fix it
- Things like this have been happening for decades -- Apollo 11LEM computer crashed during lunar descent
- The broad view: the loss was caused by a lack of system robustness in an exceptional (unanticipated) situation
Our research goal: improved system robustness