PSI - Issue 64

Ge-Wei Chen et al. / Procedia Structural Integrity 64 (2024) 724–731 Ge-Wei Chen, Xinghua Chen, Piotr Omenzetter / Structural Integrity Procedia 00 (2019) 000–000

725

2

be excited strongly enough on purpose. However, OMA is also convenient since it normally presents nil or minimal disruption to normal traffic. Thus, OMA was used in testing of short to medium-span bridges, including suspension structures (Ren et al., 2004), trusses (Debnath et al., 2012) and arches (Magalhães et al., 2012). Nevertheless, the ambient excitation is rarely pure white noise, violating the principal assumption of OMA system identification techniques. Instead, AVT has to work under variable, unknown and often inadequate excitation amplitudes, directions, durations and frequency contents. For instance, the ambient excitation may be weak and/or narrow-banded. Missing modes are commonly reported in OMA. Brownjohn et al. (1992) experienced difficulty in identifying the lateral modes of a suspension bridge, Alwash et al. (2009) could not discern resonant responses at the fundamental frequency from background vibration in a three-span, 100.5-m long bridge, whereas Chen et al. (2017) failed to extract the fundamental lateral and several high-order vertical modes of an 11-span concrete off-ramp bridge. Such incomplete outcomes can have profound negative effects on damage detection and performance assessment. In response to these challenges, hybrid vibration testing (HVT) emerged, where additional artificial excitation of moderate strength is applied to work along ambient excitation and system identification utilizes the response to the combined forcing (Reynders and De Roeck 2008; Shabbir and Omenzetter, 2008). HVT does not sacrifice the main advantage of AVT as it only uses small excitation devices when compared to those needed in the traditional experimental modal analysis, because the HVT artificial forces need only be similar to the ambient forces. Reynders et al. adopted HVT for the modal system identification of the Z24 bridge, a steel arch footbridge and a concrete stressed-ribbon footbridge, reporting promising results (Reynders et al., 2010a,b). HVT-based FEM updating and bridge structural damage identification became better-posed and allowed detecting inaccuracies in the FEM. Nevertheless, the previous HVT studies were restricted to system identification methods combining deterministic and stochastic subspace identification, which required measuring the artificial inputs. This will be inconvenient in challenging in-situ conditions. Thus, HVT combined with output-only identification techniques requires exploration and the present study just does that, combining the advantages of both AVT (no requirement for input measurement) and increased excitation levels provided by HVT utilizing only small, portable actuators. A wide range of OMA or output-only system identification techniques have been developed to this end, including peak-picking, frequency domain decomposition, maximum likelihood identification, Ibrahim’s time domain method, least-squares complex exponential method or data-driven stochastic subspace identification (SSI-data) algorithm (van Overschee and de Moor, 1994). The SSI-data algorithm proved robust in noise-contaminated testing environments and capable of yielding high quality modal identification results but suffers from large computational effort and long processing times (He et al., 2009; Chen et al., 2017). Other contemporary OMA techniques include the auto-regressive (AR) time series method (Pakzad, et al. 2011) and the eigensystem realization algorithm with observer/Kalman identification (ERA-OKID), which exhibit computational efficiency and identification accuracy (Phan et al., 2018). The parametric AR representations of measured responses can easily be converted into the natural frequencies, damping ratios and mode shapes. Multivariate AR models have been applied to describe the dynamics of structures and infer modal parameters from ambient vibrations, including industrial structures (Vu et al. 2013), buildings (Kim et al. 2016) or cable stayed bridges (Park et al. 2016). However, the AR and ERA-OKID reliability and computational efficiency when applied to full-scale, multiple-span bridges have not yet been expensively investigated or compared to the better-known SSI-data algorithm. This is then the extended focus of this paper. Two output-only time-domain algorithms, AR and ERA-OKID, were employed to recover modal frequencies, damping ratios and vibration shapes from the HVT vibration response data, with the SSI-data algorithm also applied to benchmark the AR and ERA-OKID performance. Modal identification results were also obtained from pure AVT to observe any improvements provided by HVT. A comparison between experimental and numerical (FEM) modal parameters was conducted for further insights into the modal identifiability and identification reliability from HVT. 2. Bridge structure and finite element model The Nelson St. off-ramp bridge (Fig. 1) is located in Auckland, New Zealand and is a part of a junction of three major motorways. Its total length is 272 m and it has 11 post-tensioned concrete spans. The bridge has an appreciable spatial curvature. The main span is 40 m long, whereas the other spans vary between 18 m and 26 m. The superstructure comprises a total of 137 precast post-tensioned single-cell box girder segments that were delivered to the site. The girder cross-sections are of two heights, namely, 1.73 m and 1.09 m. The solid octagonal piers have