Collapse Fragility Curves Development with Considering of Modeling Uncertainties Using LHS Simulation and Artificial Neural Network

Abstract

Collapse performance evaluation of structures has been a concern for researchers due to its complexity and uncertainty in modeling and simulation. Concentrate plastic hinges are best candidates for modeling collapse behavior of structures. Collapse fragility curves are affected by various sources of uncertainty. Existing uncertainties in modified Ibarra and Krawinkler moment-rotation model for concrete moment frame buildings were investigated in this paper. LHS simulation method was used to generate random variables considering the correlation among modeling uncertainties in one component and two structural components. Collapse responses including mean collapse capacity and standard deviation were obtained for each simulation by generating random samples for uncertainties using incremental dynamic analysis (IDA). As much effort is needed for implementation of IDA, MLP artificial neural networks, GMDH artificial neural network and response surface method were used to estimate and anticipate the collapse behavior of the structure. Results show that using above methods will lead to high accuracy anticipations with an error of less than 10% for GMDH neural network and an error of less than 7% for MLP and response surface methods.

Keywords

• collapse fragility curves
• modeling uncertainties
• lhs simulation
• artificial neural network
• response surface method