| dc.description.abstract | Modelling of wind loads on long-span suspension bridges is a challenging topic, despite the successful construction of a number of long suspension bridges globally since the early 19th century. Several engineering theories exist that attempt to describe the wind loading and the corresponding bridge response that are continuously being refined. For that, it is essential to observe characteristics of natural wind at a bridge site simultaneously with the wind-induced bridge response, and examine the link between the two with in-situ measurements in the field. The University of Stavanger’s wind engineering team have been continuously collecting wind and bridge vibrations measurement data set since 2013 on The Lysefjord Bridge, a 639 metre long suspension bridge located in South Western Norway.
Six random days were selected to use for analysis. The cross-correlation of the wind speed was studied to determine how distance, frequency and wind speed affect the correlation between winds measured at two separate locations. Secondly, the bridge response was analysed, the bridge acceleration response was calculated based on the input wind speed using a comprehensive model that employed Buffeting Theory, Modal Analysis and Numerical Integration, the results were then compared to the measured acceleration data. The standard deviation of 10-minute acceleration samples were then compared to the 10-minute horizontal mean wind speed to determine the relationship.
Two days of data were added to the study, a wind storm on August 9, 2014 which knocked over a caravan driving over the Lysefjord Bridge and a magnitude 4.6 earthquake experienced on June 30, 2017. The speed and acceleration responses were analysed and compared to the results from the six random days.
Lastly, due to the unprecedented ability of computers to collect and process new data, machine learning is a growing technology that can be applied to countless fields to interpret data. Using the data from the Lysefjord Bridge, attempts were made to model the acceleration response using MATLAB’s machine-learning regression algorithms based on the wind and acceleration data. These results were extrapolated to predict the acceleration response at higher wind speeds which was compared with the measured acceleration data.
The results from the regression algorithms were also compared to results from the more traditional curve-fitting approach using MATLAB’s curve-fitting application. | |
