The challenges of accurately identifying the dynamic characteristics of bridge structures from ambient vibration responses persist because, unlike in the ideal laboratory environment, elevated levels of noise in data are unavoidable at any in-situ testing site and may have detrimental effects on the modal parameter identification process and results. This is especially true for vibration tests conducted under weak ambient excitation sources resulting in poorer signal-to-noise ratios (SNRs). This paper presents an investigation into the feasibility and reliability of modal identification using operational modal analysis (OMA) techniques under such weak excitation circumstances and with responses measured by inexpensive stand-alone accelerometers/data recorders. An eleven-span concrete motorway off-ramp bridge, closed to traffic, was excited only by ground vibrations generated by traffic on the motorway passing underneath the bridge as well as on nearby motorway on- and off-ramps, weak winds, and possible micro tremors. A high spatial resolution of measuring points on the bridge deck was used to collect vibration responses. Three output-only modal parameter identification algorithms were utilised to extract the modal properties, namely the peak picking (PP), the frequency domain decomposition (FDD) and the data driven stochastic subspace identification (SSI) method. Nine lateral and three vertical modal frequencies below 10Hz could be identified despite the weakness of the environmental excitations and noise in sensors. The identified experimental natural frequencies were stable, damping ratios, however, had a marked scatter. A comparison with the results of a numerical modal analysis using a finite element model revealed, however, that several higher order vertical modes were missing from the experimental results altogether, and some of the OMA methods missed the fundamental lateral mode. Overall the PP method was the most successful in finding the largest number of frequencies but the SSI method yielded the highest quality mode shapes. The SSI method is, however, computationally more expensive that the remaining two methods. For quick, preliminary results, the PP and FDD methods can still be useful and detailed analyses could use SSI and FDD. Overall, the study argues that output-only system identification can provide useful quantitative insights into the modal properties of stiff bridges even under weak environmental excitations, or poorer SNRs, but its limitations need to be acknowledged.
Read full abstract