In line with the response of Athmane Tajine, we can mention several incidents of a rather severe nature due to "beasts" computer problems (which are all human faults and could have been avoided, but no responsibility has ever been retained because in all cases the industrial processes have been particularly cured and followed).
Explosion of the Ariane 5 rocket (1996, Kourou, French Guiana)
On June 4, 1996, the Ariane rocket exploded after only 36.7 seconds of flight, representing a loss of more than 500 million dollars.
The error is attributed by the Inquiry Committee [1] to the inertial reference system, which crashed after the lifting of an untreated exception, related to the conversion of a floating-point number (represented on 64 bits ) in integer (represented on 16 bits) - the variable concerned, in the ADA code, stored the value of the horizontal velocity of the rocket relative to the firing point. The code had been directly imported from an Ariane 4 module, which had a rather different take-off trajectory, and in particular never encountered too high horizontal speeds (in practice less than 32768 = 216
). This value was exceeded by Ariane 5, which resulted in an overflow.
The autopilot continued to receive totally erroneous values, and steered the engine goutiers to make a violent course correction. Dynamic forces shredded the rocket and self-destruction followed almost instantly.
Some variables were protected against this problem (the exception was caught) but three variables - considered critical - were not. Nothing explains this choice from a conceptual point of view, and it is clearly a fault.
Patriot Missile (1991, Gulf War)
In February 1991, in Dharan, Saudi Arabia, an American Patriot missile failed to intercept an Iraqi Scud missile, which eventually resulted in 28 American (military) casualties. The investigation reveals that the malfunction is due to a rounding problem resulting in an error in the calculation of travel time.
The numbers were represented on 24 bits as a fixed point, while the tick of the internal clock was one-tenth of a second. But the 1/10 writing in binary is infinite:
and the on-board computer rounded up to 24 digits, which accumulated a slight error in the countdown of time ... over time.
During the attack, the system had been running for a hundred hours, resulting in a rounding error of - still - 0.34 seconds. Over the same time interval, the Scud missile had time to cover an additional 500 meters, which is why the US missile missed its target.
The destruction of the Mars Climate Orbiter spacecraft.
The cause of this destruction is so stupid that for me it is one of the biggest scientific accidents.
In 1998, NASA launched a space probe towards Mars. This probe was to study the climate on Mars. It was to provide valuable information on the latter including the water cycles of the red planet.
In order to reduce the mission's budget, NASA outsources part of the project to a company, Lockheed Martin Astronautics. The probe must return and remain in orbit around Mars, and therefore follow a very precise trajectory.
During the journey between the earth and Mars, the trajectory is slightly corrected by thrusters so that the approach is made at the right distance (between the probe and Mars).
Finally, that's what our dear scientists thought ...
In fact, the instructions given to the probe were false.
The trajectory calculations were performed by NASA software, which used the international unit system, including the Kilogram. But the engineers did not know that the information transmitted by the probe, via the software of the subcontractor, was in book.
The confusion between the two units falsified the calculation of the trajectory, the probe was too close to Mars and flamed like a shooting star.
Here's how a stupid mistake of units turned a few hundred million dollars into ashes.