Speaker
Description
Abstract. Torsional responses in suspension bridges are of paramount importance to bridge stability and safety under strong wind conditions. Understanding and forecasting these responses can significantly aid bridge owners to ensure the health and integrity of their assets and avoid inappropriate and catastrophic consequences of aerodynamic instability and failure. This study proposes a supervised predictive approach based on dual applications of random forests (RF). The first application is related to feature selection by RF that aims at choosing the most appropriate wind features. The second application pertains to torsional response forecasting by training a RF regression model and then predicting unseen response samples under new wind features. Accordingly, the major contribution of this study is to suggest the dual applications of RF to wind-induced vibration of suspension bridges. Measured wind data (speed and direction) and acceleration responses from an anemometer and two accelerometers installed at the midspan of a suspension bridge are used to evaluate the proposed solution. Results demonstrate that the dual integration of RF for feature selection and data forecasting makes an effective and reliable predictive model for forecasting the bridge torsional response under windstorms.
