Surrogate modeling of structural seismic response using probabilistic learning on manifolds

Nonlinear response history analysis (NLRHA) is generally considered to be a reliable and robust method assess the seismic performance of buildings under strong ground motions. While NLRHA fairly straightforward evaluate individual structures for select set motions at specific building site, it becomes less practical performing large numbers analyses either (1) multiple models alternative design realizations with site-specific motions, or (2) archetype sites sets In this regard, surrogate offer an running repeated NLRHAs variable paper, recently developed modeling technique, called probabilistic learning on manifolds (PLoM), presented estimate structural response. Essentially, PLoM provides efficient stochastic model develop mappings between random variables, which can then used efficiently responses systems variations in design/modeling parameters motion characteristics. The algorithm introduced two case studies 12-story estimating probability distributions responses. first example focuses mapping multidegree-of-freedom its peak story drift acceleration second applies technique both examples, training data are generated orthogonal input parameter grids, test prescribed statistical distributions. Validation performed examine accuracy efficiency models. Overall, examples show good agreement estimates verification sets. Moreover, contrast other common techniques, able preserve correlation structure Parametric conducted understand influence different tuning prediction accuracy.