Optimum retrofit strategy of FRP column jacketing system for non-ductile RC building frames using artificial neural network and genetic algorithm hybrid approach
Jiuk Shin, Sang-Ki Park
Abstract
Existing reinforced concrete building structures have seismic and blast vulnerabilities due to their seismically deficient details. Such structural vulnerabilities can be mitigated using a fiber-reinforced polymer jacketing system (circular shape of prefabricated jacket shell and grout materials infilling annual spaces), which provides additional confining pressure to existing columns. To find the optimum retrofit scheme, a repeated procedure for designing, modeling, and simulating the retrofitted structure can be time-consuming. This paper proposed a rapid decision-making tool developed using a hybrid machine-learning technique, which can immediately derive optimum retrofit schemes without the laborious nature of manually repeated procedures. The hybrid technique consists of an artificial neural network for rapidly generating structural responses and a genetic algorithm for optimizing retrofit details (jacket strength and thickness mainly related to confinement; and grout strength and inner diameter of columns related to stiffness) under confinement and stiffness parameters. The machine-learning based tool optimized the retrofit details within target performance levels through maximizing the confinement ratio and minimizing the stiffness ratio, and it derived acceptable ranges of seismic loads. Based on the investigation, the geometric conditions-related stiffness parameters within a low confinement level were increased rather than increases in the confinement parameters. However, to extend acceptable ranges of seismic and blast hazard levels, the retrofit details were optimized with maximizing the confinement parameters.