There have been many incidents where bad numerical computing practices have led to massive financial and structural losses. Poor handling of rounding errors led to the Patriot Missile failure in Dharan Saudi Arabia on February 25 1991, and an integer overflow led to the explosion of the Ariane rocket just after lift-off on its maiden voyage off French Guiana on June 4 1996.

Similarly, on August 23 1991, the Sleipner A platform, a concrete deep water structure which produces oil and gas in the North Sea, sank during a controlled ballasting test to check for a minor leakage in Gandsfjorden outside Stavanger, Norway. The collapse caused a seismic even registering 3.0 on the Richter scale. An investigation conducted by the owner of the platform, Statoil, a Norwegian oil company, revealed that the collapse was caused by a combination of a serious error in the finite element analysis, and insufficient anchorage of the reinforcement in a critical zone. It was estimated that the financial loss from the accident was 250 million US dollars.

The platfom consists of two units, the hull and the deck. The hull is a gravity base consisting support pilings and concrete chambers, above which three or four shafts extend out to support the deck. Investigations showed that a wall of one of the tricells of the hull had failed, leading to the flooding of the structure and its eventual collapse. This was due to errors in the Finite Element analysis of the platform - insufficient discretization and poor geometrical shoping of the elements caused the algorithm to underestimate the stresses in the structure by about 50%, and led to a design where the cell walls were too thin in critical areas.

This incident highlights the need for us to perform manual sanity checks on the calculations and results produced by computer algorithms. The development of complex and highly automated computer algorithms and programs have led to many engineers and structural designers blindly making use of such technology without checking and making sure that the results obtained are correct. Especially with the high costs and stakes involved in such projects, it is probably critical and definitely more financially sound to perform such checks to avoid such costly accidents.

Sources

http://www.ima.umn.edu/~arnold/disasters/sleipner.html

http://books.google.com/books?id=INZNrLBn4cUC&pg=PA246&lpg=PA246&dq=23+august+1991+sleipner&source=web&ots=XsxXiTbS6m&sig=HxcggqHS60vjFGMHtt743v_lIcM#PPA246,M1

http://home.versatel.nl/the_sims/rig/sleipnera.htm