Sequential damage detection approaches for beams using time-modal features and artificial neural networks

In this study, sequential approaches for damage detection in beams using time-modal features and artificial neural networks are proposed. The scheme of the sequential approaches mainly consists of two phases: time-domain damage monitoring and modal-domain damage estimation. In the first phase, an acceleration-based neural networks (ABNN) algorithm is designed to monitor the occurrence of damage in a structure by using cross-covariance functions of acceleration signals measured from two different sensors. By using the acceleration feature, the ABNN is trained for potential damage scenarios and loading patterns which are unknown. In the second phase, a modal feature-based neural networks (MBNN) algorithm is designed to estimate the location and severity of damage in the structure by using mode shapes and modal strain energies. By using the modal feature, the MBNN is trained for potential damage scenarios. The feasibility and the practicality of the proposed methodology are evaluated from numerical tests on simply supported beams and also from laboratory tests on free–free beams.