Keywords: structural reliability, neural networks, support vector machines, Monte Carlo simulation, entropy, statistical learning, particle swarm optimization.

A general method for using learning classifiers, such as neural networks, support vector machines and others, for structural reliability analysis is proposed. The method comprises two parts: (a) a technique for generating the learning population and (b) a method for selecting the Monte Carlo population necessary for estimating the average probability of failure using the classifier. For the first part, the generation of the learning population is accomplished by solving a simple unconstrained optimization problem in terms of the limit state function. With respect to the second part, it is recalled that the estimate of the failure probability by means of Monte Carlo simulation is a random variable as such, so that the conventional use of an arbitrary random population may yield results deviated from the average value. To overcome this it is shown that a preprocessing of candidate populations according to their entropy offers a population yielding a failure probability estimate close to the average computed among a large number of classifier exploitation sets. The chapter ends with a discussion about the dimensionality problems in common soft-computing classifiers used in structural reliability.

purchase the full-text of this chapter (price £20)

go to the previous chapter
go to the next chapter
return to the table of contents
return to the book description
purchase this book (price £88 +P&P)