A data-driven approach to predict the natural frequencies of similar bridges

Selected results collected in the dynamic monitoring of two similar multi-span steel-concrete composite bridges are reported in the paper. The investigated bridges consists of 4 spans and belong to a motorway serving several provinces of the Lombardy region. After a concise description of the bridges and the monitoring systems, details are given on the dynamic characteristics identified in about 16 months of monitoring: although the differences in the span lengths and transverse cross-sections of the two bridges induce different natural frequencies, an excellent correspondence is found in terms of mode shapes of the lower 12 modes for the two monitored infrastructures. The almost linear dependence on ambient temperature that is exhibited by the identified natural frequencies of both bridges as well as one long interruption in data acquisition that occurred on one of the two infrastructures, suggested to use (after an appropriate training period) the knowledge available on the natural frequencies of one bridge to predict the unobserved frequencies of the other bridge. In addition, when both monitoring systems are active, the residuals between identified and predicted frequencies should be adopted to drive a novelty detection approach, with this task not requiring any minimization of operational and environmental effects.