Machine learning-based design of a seismic retrofit frame with spring-rotational friction dampers

In this study a machine learning-based design procedure for a new seismic retrofit system is presented and its seismic retrofit capability is validated by intensive seismic analyses. The retrofit system consists of a steel frame with rotational friction dampers (RFD) at beam-column joints and linear springs at the corners providing both energy dissipation and self-centering capabilities to existing structures. The performance-based seismic design procedure of the spring-rotational friction damper (SRFD) retrofit system is developed using a genetic algorithm (GA) and an artificial neural network (ANN). The performance of the presented retrofit system and its optimum design procedure is evaluated using case study models for multi-limit states by investigating seismic fragilities, life-cycle cost (LCC), and seismic Resilience Index (RI) before and after the retrofit. According to the analysis results, the SRFD retrofit system proved to be effective in decreasing story drifts, seismic fragility, and LCC of the retrofitted structures significantly. The retrofitted models turn out to be more resilient and recover better than the un-retrofitted models after earthquakes. The developed performance based design procedure is proved to be effective in seismic retrofit of case study structures to satisfy multi-level design objectives.