Inverse design of isolated buildings given desired collapse probability via surrogate model

While isolated structures offer enhanced, predicable behavior under typical ground motions, their behavior under extreme motions is complex and may not reach targeted goals as currently outlined in various codes. With isolation, the designer should be able to target their desired level of collapse, however, it is infeasible to rerun an incremental dynamic analysis with each change in design parameters. Therefore, it is sought to solve the inverse design problem using Gaussian process surrogate modeling to associate the probability of building limit states with the input design parameters. A large data base is generated using random design values with automated design of triple friction pendulum isolated moment frames and ground motion scaling. The analysis is conducted in OpenSeesPy. Defining failure as exceedance of story drift limits, failure quantification can be placed on each model analysis. The database is used to build Gaussian process surrogate models from which the probability of collapse is generated given a set of design parameters. Given a target collapse probability, an admissible design space is identified from which a final design is selected based on a simple cost optimization strategy. The required designs for various probabilities of collapse are compared and validated.