ABSTRACT
Base isolation systems are often chosen in structural engineering to outperform conventional designs, allowing designers to achieve high performance goals. However, achieving specific performance targets requires numerous high-fidelity analyses and often involves iterative design. To address this, a database of diverse structural designs under earthquake inputs is constructed, and probabilistic loss and downtime estimation is carried out. Several machine learning models are utilized to fit the design parameters to their performance, allowing for prediction over a wide space of new design combinations. The prediction models find that accurate cost estimation relies on classification of impact against the moat wall, which is strongly predicted by the moat capacity of the isolators. The developed model is used to for an inverse design procedure in which acceptable design regions are attained by defining acceptable limits for repair cost, downtime, etc, and the resulting design is verified to outperform a typical code compliant design.
