B1364
Title: The application of black-box modeling techniques to remove environmental influences on vibration monitoring data
Authors: Kristof Maes - KU Leuven (Belgium) [presenting]
Geert Lombaert - KU Leuven (Belgium)
Abstract: Railway bridge KW51 in Leuven, Belgium, has been continuously monitored since October 2018. The modal characteristics (natural frequencies and modes shapes) are tracked over time with the aim of detecting changes in the structure that could potentially be attributed to damage. During the monitoring period, the railway bridge has been retrofitted in order to resolve a construction error. The retrofit results in data for two distinct states of the structure, which makes this case study particularly relevant within the field of structural health monitoring. The focus is on removing the effects of environmental conditions, such as temperature, which affect the modal characteristics of the structure and, therefore, may lead to false-positive or false-negative damage detection. A comparison is made between standard linear regression and robust principal component analysis (PCA). In order to assess the success rate of these techniques, a receiver operating characteristic (ROC) curve analysis is performed, considering the actual retrofit as well as a number of more subtle structural changes, which are modelled using a detailed finite element model of the structure. The state transition can be observed for the actual retrofit as well as for smaller structural modifications that result in relatively small natural frequency shifts.
