Vibration-based damage localization in Ting Kau bridge using probabilistic neural network

This paper presents a study of using the probabilistic neural network (PNN) to identify the damage type and region in the cable-stayed Ting Kau Bridge from the simulated noisy modal data. The essence of the PNN for damage type and region identification is to judge the pattern class of damage types and regions to which the test vectors of unknown source should belong. In the present study, a total of 17 pattern classes are defined for the Ting Kau Bridge to depict various damage types and different damage locations. The simulated damage cases involve the damage occurring at the main stay cables, longitudinal stabilizing cables, transverse stabilizing cables, main girders, cross-girders, and bearings. The characteristic ensembles for each pattern class (training samples) are obtained by computing the natural frequency change ratios when incurring the corresponding damage in a validated finite element model of the bridge and then corrupting the analytical frequencies with random noise. The testing samples for damage localization exercises are obtained in a similar way except that the damage is incurred at a different location of the same region for each pattern class. The simulation study results show that when the first twenty natural frequencies are used, the damage type and region can be identified with a high confidence even if the measured natural frequencies are polluted by the random noise with 10% standard deviation.