Neural networks application in seismic reliability and performance-based design

摘要

Seismic structural reliability analysis and performance-based design must consider the many uncertainties that arise from the earthquake ground motion, the structural geometry and material properties, the analytical models used in determining the structural response and the relationships between accumulated damage and quantifiable parameters like deformations. The approach presented here involves calculation of responses by an appropriate nonlinear dynamic analysis, followed by their representation using neural networks and the calculation of reliability for a set of specified performance requirements via importance sampling simulation. In performance-based design, an optimization procedure permits the calculation of design parameters, for specified target reliabilities, using a neural network representation of reliabilities achieved for different combinations of the design parameters. Two application examples are shown: a wood shear wall common in housing applications, and a frame for a tall reinforced concrete building.