ABSTRACT
Bridges are getting older and older in Germany. This leads to a high demand of maintenance in the future. Stuctural health monitoring can be a helpful tool for bridge assessment in identifying bridge damage in the early stages. Our goal is a system that is non-invasive, mobile and can be retrofitted on existing bridges under ongoing operation of traffic. The conventional methods of bridge assessment and monitoring use strain measurements and acceleration sensors and are based on modal properties – such as natural frequencies, modal damping and displacement modes or curvature/strain modes – which are prone to environmental changes such as temperature and vehicles. Recent papers introduce features based on the curvature influence line. Some of the introduced features are based on ratios and single vehicle crosses. The advantage of ratio-based features is the reduction of the unknown load. The results show promising results for strain based signals. The question now arises how well the ratio-based features for strain signals can be transferred to the displacement signals measured by an interferometric radar. The advantage of using radar over strain sensors is that the radar is non-invasive and mobile. We compare here the performance of ratio-based features using the displacement influence line with the same features using the curvature/strain influence line for damage detection. In addition to the ratio-based features, we investigate for the comparison other features like peak position and maximum value. We assess the performance of the features with supervised machine learning with various combinations of damage position and damage severity. The considerations are made with an analytical stepped continuous Euler-Bernoulli beam.
