Keywords: neural networks, reliability analysis, nonlinear dynamic analysis, fragility analysis.

Extreme earthquake events may have a very low probability of occurrence, but when this happens it produces extensive damage to structural systems. Therefore, it is essential to establish reliable procedures that are capable of predicting the seismic risk of real world structures. These procedures can only be created within the context of a probabilistic safety analysis (PSA) approach, which provides a rational framework for taking into account the various sources of uncertainty that may influence the structural performance and/or the seismic hazard. Seismic fragility analysis, which provides measurements of the safety margins of a structural system above specified hazard levels, is the core of PSA. On the other hand, the problem of simulating non-Gaussian stochastic processes and fields has gained considerable interest in stochastic computational mechanics over the past years. This can be attributed to the fact that several quantities arising in practical engineering problems (e.g. material and geometric structural properties, soil properties in geotechnical engineering applications, etc.) exhibit non-Gaussian probabilistic characteristics. Especially the simulation of highly skewed and narrow-banded fields is well recognized nowadays as a benchmark that reveals the limitations of the existing simulation methods.

The theory and methods of structural reliability have been developed significantly during the last twenty years and have been documented in an increasing number of publications. In this work the probabilistic safety analysis of framed structures under seismic loading conditions is investigated. Both randomness of ground motion excitation (that influences the seismic demand level) and material properties (that affect the structural capacity) are taken into consideration. The assessment of the bearing capacity of framed structures, in terms of maximum interstorey drift, is determined via non-linear time history analysis. The PSA using Monte-Carlo simulation (MCS) and non-linear time history analysis results in a highly computationally intensive problem. In order to reduce the computational cost, efficient neural networks (NN) metamodels are employed. For the training of the NN a number of intensity measures (IM) are used in order to accurately predict maximum interstorey drift values. The IMs adopted in the present study can be classified either as seismic record dependent only, or as both structure and record dependent. Via the proposed PSA procedure fragility curves are obtained for different hazard levels. In addition the probability of a structure's failure is derived as a limit state function of seismic intensity.

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)