A modal survey was performed on eight posttensioned reinforced concrete beams removed during the demolition of a highway bridge in the United Kingdom. The aim of the study was to continue investigations into using vibration data for the structural health monitoring of bridges. The use of modal data to distinguish between beams with different damage levels and between modified and unmodified beams was inconclusive. However, there were clear differences in the shape of the measured frequency response functions (FRFs), which suggested that this might be a better classifier. The method of principal component analysis was used to reduce the FRF data dimensionality, and when this was done, it was possible to clearly distinguish between damaged and modified beams using only a two-dimensional feature vector. To distinguish between beams with different damage levels, feature vectors with more components were used in conjunction with a generalized linear artificial neural network. This approach successfully classified the beams according to damage level, although more beams need to be tested to improve the reliability and general applicability of the method.
Abstract