Output-only entropy-based damage detection using transmissibility function

This paper presents an output-only structural damage identification methodology by combining the concept of information entropy and transmissibility function. Information entropy quantifies complexity change of structural behavior induced by damage. However, many of the real-life structures are under random input excitations that impose complexity to the structural response. The complexity change induced by the random input may be misinterpreted as damage; thus, the inverse transmissibility function (ITF) is used to eliminate the influence of random input excitation on the complexity change. Damage index is introduced by calculating the entropy of the ITF, and damage identification ability of the ITF-based damage index is compared to the acceleration and impulse response function. The proposed methodology is studied on a numerical and a standard experimental dataset to identify damage presence, type and severity. A multi-layer perceptron neural network is implemented to classify the damaged samples in the experimental dataset, and different damage indices are compared according to the classification results. The case studies validated the effectiveness and robustness of this novel damage identification methodology.