Stochastic determination of changes in sensor-based modal frequencies

Abstract

Changes in main modal frequencies of structures due to damage (or repairs) are difficult to determine with accuracy because measurement-derived structural parameters are usually highly spread and obscured by non-random noise (bias). A new simple stochastic method to compute such changes is presented. This method is based on a constrained cross-correlation analysis of redundant spectra of free attenuating oscillations during two periods of measurements. Bias (non-random noise) characterising sensor-derived data is randomised, and this permits precise estimates of changes in modal frequencies and of the corresponding confidence intervals. The proposed method was applied in a pedestrian bridge in Greece and permitted to document statistically significant changes in its first vertical modal frequency only during a certain interval for which there is independent evidence of damage. The method proved functional at a level of 0·1 Hz, covering also the possibility of false and missed alarms. The overall approach is complementary to formal structural identification techniques and may prove useful for structural health monitoring in the perspective of resilient and functional modern infrastructure.