Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
With the overall goal of reducing the numerical burden of uncertainty quantification involved in seismic risk assessments, this paper examines opportunities for assessment of seismic collapse capacity and risk using data-driven surrogates as a complement to physics-based response history analyses. Specifically, a methodology is proposed for computing collapse fragilities for generic sets of earthquake ground motions (i.e., not hazard-consistent sets) wherein collapse capacities for only a portion of the set are computed with incremental dynamic analysis (IDA) and used as the training data for surrogate modeling. The capacities for the remaining motions, referred to as the test set, are then estimated using the recently introduced data-driven collapse classifier (D2C2) and the automated collapse data constructor (ACDC) technique. These predictions of the collapse capacities of the test set are combined with the training data obtained using IDA to yield the generic collapse fragility of the entire input set of motions. Anti-clustering is used to split the ground motions into the training and test sets to make them as similar as possible to each other while maximizing the difference between ground motions within each set. Scalar intensity measures are used as inputs for the data-driven surrogate. The methodology is tested in a case study using steel moment resisting frames ranging from 4 to 20 stories and the FEMA P695 Far Field ground motion set. The results demonstrate the feasibility of the proposed methodology as well as the utility of “small data” machine learning approaches for seismic collapse risk assessments.
, , ,