Seismic performance prediction of steel structures using multiple intensity measures

The prediction of performance of structures in seismically active regions has been traditionally based on a single scalar ground motion parameter (GMP) or intensity measure (IM). However, it is widely accepted that the performance of a structure subject to ground shaking is probably dependent on more than a single parameter. Therefore, performance predictions based on a scalar ground motion parameter are subject to a great deal of variability, which in turn requires greater effort to obtain results with a reasonable level of confidence. It is expected that by including more information about the ground motion in our set of predictors we might be able to predict structural performance with smaller variability, or equivalently, require less effort to predict performance with the same level of confidence. The objective of this study is to investigate whether vector combinations of IMs might correlate better with structural performance than use of any parameter singly. A total of 140 recorded ground motions are considered in this study. These motions include a "near-source" set of 70 motions recorded at distances less than 16 km from the source as well as an "ordinary set" of 70 motions recorded at distances greater than 16 km. Results of nonlinear dynamic analyses of 3-, 9-, and 20-story steel moment-resisting frame buildings designed by practicing engineers for the Los Angeles region as part of the SAC steel project, subjected to the selected ground motion records, will be used to evaluate the predictive power of alternative vector-valued IM sets. Because knowledge of more than a single structural performance measure can help reduce uncertainty in predicting losses and damage levels, vectors of such performance measures (e.g., various drift ratios) resulting from dynamic analyses with the selected ground motions will be studied for their dependence on the selected GMPs. The degree of correlation between the alternative vector GMP sets and the structural performance vectors will be evaluated via multivariate multiple linear regression. Such regressions are improvements over existing procedures that either only consider individual structural performance measures separately or employ a single GMP. Example loss estimates based on selected vectors of structural performance measures are presented to demonstrate how the regression studies may be used in practice.