Neural Networks to Optimize Design Parameters of Bridges Isolated with Double Concave Friction Pendulum

The use of machine learning techniques in the field of numerical optimization represents a promising strategy to approach dynamic problems characterized by strong non-linearities and a high number of parameters. Seismic vibration control is undoubtedly one of the fields that can benefit most from this approach, especially when aiming at studying the effectiveness of innovative techniques on a large scale and the definition of new design guidelines. In this context, this study aims to model the seismic behavior of Double Concave Friction Pendulum (DCFP) isolated deck bridges through a trained artificial neural network (ANN). Specifically, the investigation employs several data representing the seismic response to a significant number of synthetic seismic excitations related to different soil conditions and PGV/PGA ratio values. A comprehensive analysis was conducted by using the ANN model, leading to empirical considerations regarding the optimal design of the DCFP device on varying the dynamic characteristics of both the structure and the input excitation.