Failures and Reliability of Large Projects
Learning from failures has long been a central part of geotechnical engineering. Although there are many types of geotechnical failures, dam failures are especially critical and, therefore, tend to be investigated carefully, making it possible to obtain some idea of what happened. The dam failures at South Fork, Saint Francis, Malpasset, Vaiont, Baldwin Hills, Teton, and the slope failure at Aberfan demonstrate the consequences of failure to address geotechnical and geological issues adequately. However, they also show that an inadequate safety culture, overconfidence, and confused management are often more important than purely technical issues. Failures in other areas (the nuclear power plants at Three Mile Island, Chernobyl, and Fukushima; the space shuttles Challenger and Columbia; the events surrounding Hurricane Katrina; and the Deepwater Horizon oil blowout) have their own characteristics peculiar to their industries, but organizational features and mental attitudes seem to be present regardless of the particular technologies involved. How can the engineering profession and the public deal with the potential for failure?
Probabilistically based reliability methods make it possible to go beyond a simple estimate of settlement or factor of safety to provide values for the probability of failure and the uncertainty in projected performance. Once almost exclusively the subject of research studies, they are now finding increased application in geotechnical engineering. The computational methods are fairly well known, but we must also understand the sources of geotechnical uncertainty and how they affect the analytical results. In many ways the most important issue is how reliability results can be understood in the context of other risks and how they can be presented to others. Reliability methods can also incorporate the estimated uncertainty due to inadequate human performance.
The uses of reliability methods go far beyond the simple calculation of the probability of failure. One direct result of reliability analysis is the ability to identify the parameters whose uncertainty contribute most strongly to the probability of failure. This then identified where improved exploration and testing will have the most effect. Similarly, the reliability studies provide guidance on what field observations will be most useful. A second benefit of reliability studies is that, when the reliability of the project is not considered adequate, they indicate whether it will be more effective to revise the configuration of the project or to reduce uncertainty by more extensive exploration, testing, and analysis. A third improvement made possible by reliability analysis is that tools such as the F-N plot can be used to demonstrate the relative risks of the project compared to other risks and standards. All of these types of results can be displayed in formats that are readily understood by decision makers.
The underlying message gleaned from an examination of failures and reliability methods is that we have greatly improved our ability to deal rationally with uncertainty but we must still proceed cautiously in a world where the unexpected can and often does happen.