Performance-based earthquake engineering (PBEE) is a probabilistic approach to quantify buildings performance against earthquakes in terms of decision metrics such as repair cost and functionality. Despite its advantages, integrating PBEE into design and assessment of infrastructure imposes significant challenges on computational feasibility, scaling the framework across a high-dimensional problem space, and accrual of various aleatoric and epistemic uncertainties. This presentation discusses how machine learning (ML)-based surrogate models can tackle the challenges of computational expenses and data high-dimensionality to provide rapid, accurate, and interpretable design solutions. Illustrative examples are provided to discuss ML-based surrogate models capabilities and limitations with respect to estimating seismic losses of recently developed databases of steel and concrete frames. Possible extensions and implementations as part of a reliability-based design decision support will also be discussed.

Performance-based earthquake engineering (PBEE) is a probabilistic approach to quantify buildings performance against earthquakes in terms of decision metrics such as repair cost and functionality. Despite its advantages, integrating PBEE into design and assessment of infrastructure imposes significant challenges on computational feasibility, scaling the framework across a high-dimensional problem space, and accrual of various aleatoric and epistemic uncertainties. This presentation discusses how machine learning (ML)-based surrogate models can tackle the challenges of computational expenses and data high-dimensionality to provide rapid, accurate, and interpretable design solutions. Illustrative examples are provided to discuss ML-based surrogate models capabilities and limitations with respect to estimating seismic losses of recently developed databases of steel and concrete frames. Possible extensions and implementations as part of a reliability-based design decision support will also be discussed.