Design of isolated buildings to achieve targeted collapse limits through Gaussian process modeling

For conventional buildings, design codes are written such that minimally compliant buildings should achieve code-defined collapse probabilities under maximum considered events (MCE); however, there is no systematic methodology to ensure this is true for isolated buildings. Iterative design in which a building is designed, its collapse probability is assessed via incremental dynamic analysis (IDA), and the design is adjusted is inefficient given the number of permutations in the building-isolation system design. In an effort to find acceptable design parameters to achieve desired levels of collapse in early stages, full non-linear dynamic analysis can be bypassed through the use of data-driven predictive modeling. Gaussian process (GP) regression is utilized to fit and predict isolated building performance based on a generalized database. A sequential adaptive design of experiment (DoE) method is then used to strategically select training points to minimize the predictive variance of the GP model, and the model is then used to predict collapse probabilities for isolated structures. From here, designs that admit higher collapse probabilities than the threshold are removed, and cost optimization is performed to inversely select designs parameters such that the building will meet the targeted collapse probability under the MCE. Using this information, discussion on necessary required design parameters (e.g., moat clearance and strength reduction ratio) is given.