Fusion of structural damage identification results from different test scenarios and evaluation indices in structural health monitoring

There are many existing algorithms and damage indices that can effectively meet the engineering need in damage diagnosis. However, all of them (except those with an exact formulation) are based on certain assumptions with approximations, and their performances depend on the combination of test parameters, for example, type, number, and location of excitations and sensors; identification algorithm adopted; and environmental noise effect. Results obtained from different combinations of these test parameters are different and they may be mathematically biased or contradicting. Existing practice adopts the general indication of result for further maintenance decisions or by subjective screening the results with engineering sense. This article demonstrates the possibility to have a better design of the structural health monitoring system where the benefits of different damage indices and evaluation methodologies can be reaped via the fusion of the identified results. The scenario for discussion is the vibration-based damage localization problem. The change in mode shape in the modal domain and change in the covariance of impulse response function change in the time domain are selected as damage indices for the illustration. An exact form of the covariance of impulse response function is also proposed for this purpose. The demonstration is conducted via a simulated truss structure, an experimental beam, and 280 days valid recorded data from an in-service suspension bridge deck. Different measurements are analyzed to produce large number of identified results for reducing traffic excitation effect, the temperature effect, and so on, and to enhance the structural damage information in the final set of results. Experiences in handling the structural health monitoring data for monitoring the movement of a structural joint before and after a typhoon are also presented. The proposed strategy does not need a mathematical model of the structure, but this leads to heavy sensor requirement for a fine spatial resolution of the local damage.