ABSTRACT
Artificial neural networks (ANNs) are most often used for function approximation and for an object classification even if only incomplete and noisy data are available (Rafiq et al. 2001). In structural reliability analysis the role ofANN as a universal tool for function approximation is utilized when the limit state function under consideration is complicated and computer-time consuming, cf. Hurtado & Alvarez (2001), Gomes & Awruch (2004). Typical examples are nonlinear problems, e.g. the assessment of post-buckling strength of plates or shells. The inevitable FEM calculations of strength are carried out for suitably chosen sets of training and validation patterns. A subsequent reliability analysis then can be performed by the obtained ANN approximation of the strength function.
