Conference Agenda

It is generally accepted that engineering structures, whether bridges or aircraft, should be designed to have failure probability no higher than 10−6 per lifetime. The current probabilistic and computational predictions cannot satisfy this goal. While sophisticated probabilistic models have been developed to deal with the randomness of loads, the problem is the uncertainty of material failure, which has been relegated to empirical understrength (or capacity reduction) factors. Historically, the design equations of all design codes have been “marginal” equations, which is our name for the equations set at the lower margin of the test data cloud (depending on structure size, 25% to 40% below the data mean in the case of RC beam shear). Unfortunately, the offset of the mean and the variance of the database are not declared in the design code and have remained buried in the code committee documents. Furthermore, probabilistic modeling of the mechanics of failure process that determines the structural strength has been absent and the probability distribution function (pdf) required to extrapolate to 10−6 has been chosen arbitrarily – often as the lognormal pdf, which gives the lowest (and thus least safe) estimate of the understrength factor. This pdf, which represents the least safe assumption possible, is shown to be physically impossible for a database with one-and-the-same concrete while playing some role in a database comprising concretes of very different strengths. All of these problems have rendered the current failure probability estimates of concrete structures virtually meaningless when computational stochastic mechanics software is used. There is a looming crisis and concrete design code clarifications are urgently needed. Related to this is the previous conclusion that the blind prediction competition of a single large test, in which only the required strength of concrete is revealed (which is all that is required in the design code), has been misleading. The sine qua non of the remedy is that the offset of the data mean from the code equation and the coefficient of variations of the data set must accompany each design code equation. A realistic probability distribution (or, at least, its acceptable forms) should also be suggested. To this end, a proof that the lognormal distribution, though often used in practice, can never characterize structural failure probability. More genrerally, it is proven that no distribution with positive skewness is possible. At 10−6 this makes a difference as big as 1 : 2 to 1 : 3 in terms of the standard deviation. A complete remedy would require revising all the load and understrength factors of the design code and rescaling the design equations to the database means.