A Criticism of Statistical Methods in Probabilistic Models in Structural Reliability

by Karl Breitung, Univ of New South Wales, Kensington,

Document Type: Proceeding Paper

Part of: Probabilistic Mechanics and Structural and Geotechnical Reliability


The usual statistical models, classical and Bayesian, are not well suited for problems of structural reliability, since here we have rarely as assumed for these models a sequence of identical experiments (structures); in general the best are data from similar experiments (structures) and the focus is not on estimating parameters, but on calculating the probability of failure for one single structure. Therefore the statistical modelling in reliability should be done by empirical Bayes methods suited for the analysis of data from similar experiments. Further instead of concentrating on parameter estimation the statistical viewpoint of predictivism should be adopted, which means that the main objective of statistical inference is the prediction of further observations. Starting from a predictivistic point of view and comparing observations with the predictions of a model it will be possible to 'validate' a probabilistic model at least in the sense that it does not make wrong predictions.

Subject Headings: Structural reliability | Parameters (statistics) | Statistics | Probability | Structural models | Structural analysis | Bayesian analysis | Failure analysis

Services: Buy this book/Buy this article


Return to search