Bayesian Modal Identification using Asynchronous Ambient Data

Operational modal analysis (OMA) has been used nowadays for identifying the modal properties of structures (e.g., natural frequencies, damping ratios and mode shapes) based on the measured ambient vibration data for its high economy and efficiency. In vibration tests, multiple sensors are often used to obtain the mode shape information of the instrumented structure. Time synchronisation among the sensors/data channels is normally required by conventional modal identification techniques. In full-scale tests, this often requires additional equipment and logistics in order to obtain synchronous data. Ambient vibration tests can be conducted more flexibly and efficiently if time synchronisation is not required in OMA. Motivated by the above concerns, this thesis aims at developing Bayesian OMA approaches based on asynchronous ambient data. This thesis first investigates the characteristics of asynchronous data in OMA. Inspired by the experimental findings, a stationary stochastic model is proposed to model asynchronous data in OMA with imperfect coherence to capture the key asynchronous characteristics within suitable time scales. The theoretical properties of the power spectral density (PSD) matrix for asynchronous ambient data are also derived. Based on this model, two Bayesian OMA methods using fast Fourier transform (FFT) of asynchronous ambient data are proposed. Balancing simplicity and utility, the first method assumes zero coherence among the synchronous data groups, which leads to an efficient algorithm for determining the most probable values as well as the posterior uncertainties of modal properties. The second method assumes general coherence values among the synchronous data groups, which strictly obeys the proposed asynchronous data model. It provides a more robust means to identify the modal parameters based on asynchronous data without zero coherence approximation, although more computational efforts are needed. Synthetic, laboratory and field test data are used to verify and illustrate the asynchronous data model and these two Bayesian OMA methods. A full-scale ambient test with multiple setups is also presented, where the challenges and complications related to ambient vibration test with asynchronous data in real applications are investigated. This work is expected to gain more insights on time synchronisation problems in OMA and provide a pathway for more flexible and economical implementation of ambient vibration tests