A heuristic approach to output-only system identification under transient excitation

Output-only system identification is a very attractive technique for its implementation simplicity. However, it requires long records to validate the white-noise assumption of the excitation, mainly under transient forced vibration. Alternatively, free-vibration record segments can be selected before the identification process. This improves the accuracy, even using less data, but it requires human intervention or input recording. In the present paper, an approach is proposed for accurate system identification from short output-only records of vibration induced by transient excitation, without human intervention. The approach is based on a novel heuristic search algorithm to find free-vibration record segments, which is fully automatic and it handles the possibility of free-vibration absence. Tests with real-life data from Structural Health Monitoring (SHM) of a bridge showed that the free-vibration finding improves the accuracy of the modal parameter estimates up to ten times, as compared to using record segments starting at the response peak. The proposed approach drastically reduces the need to transmit large amounts of data, which impacts on hardware requirements of SHM implementations.