Damage severity assessment of timber bridges using frequency response functions (FRFs) and artificial neural networks (ANNs)

This paper presents a novel vibration-based technique that utilises changes in frequency response functions (FRFs) to assess advancement of damage in timber bridges. In the proposed method, damage patterns embedded in FRF data are extracted and analysed by using a combination of principal component analysis (PCA) and artificial neural network (ANN) techniques for estimation of severity levels of damage. To demonstrate the method, it is applied to a laboratory four-girder timber bridge, which is gradually inflicted with accumulative damage at different locations and severities. To extract damage features in FRFs and to compress the large size of FRF data, FRFs are transferred to the principal component space adopting PCA techniques. PCA-compressed FRF data are then used as inputs to ANNs to identify severities of damage. The excellent severity predictions obtained from the ANNs show that FRF data can potentially be very good indicators for the assessment of damage advancements in timber bridges.