MS18.9 - System Identification and Damage Detection

In Cahora Bassa dam, a 170m-high arch dam located in western Mozambique, a system for vibrations monitoring of the type SSHM (Seismic and Structural Health Monitoring) was installed in 2010. The dam, in operation since 1974, is affected by a swelling process of small/medium amplitude. Aiming to control its structural health over time, the dam was instrumented with the referred SSHM system which includes 10 uniaxial accelerometers in the crest gallery and 3 triaxial accelerometers in the dam-foundation interface. The latest results on the dynamic behaviour monitoring of the dam are presented, namely the evolution of the main natural frequencies over time and the corresponding mode shapes. Using HST statistical models for separation of effects, the evolution of frequencies over the last 12 years is analyzed in order to detect a possible global stiffness decrease induced by the swelling process. A comparison is made between HST results and numerical results from a 3DFE model of the dam-reservoir-foundation system.

When performing modal identification of structural systems, the main goal is to accurately extract the modal properties from the measured vibration data. The system identification can be carried out either in time or frequency-domain depending on the primary data used in the identification process, i.e., a time-domain free decay or frequency-domain function. In this paper a novel system identification technique formulated in the frequency domain is introduced. The technique consists of a frequency domain method entitled poly-refence Complex Frequency formulated in Modal Model (pCF-MM). The idea behind the formulation of such approach is to derive an eigenvalue problem by comparing to the frequency-domain function (e.g., the Half Spectrum (HS) or the Frequency Response Function (FRF)) evaluated at two neighbouring frequency lines. In order to illustrate its robustness, the novel pCF-MM is applied to the vibration data of a balcony specimen to estimate its first vibration modes.

The evaluation of the actual state of structural systems based on available response measurements is a relevant aspect in civil engineering. To this end, Bayesian model updating provides a sound theoretical framework that allows to include the unavoidable uncertainties arising in modeling and monitoring processes. The effective treatment of high-dimensional parameter spaces in this framework represents one of the open challenges for practical applications, especially for complex non-linear structural models. To address this issue, an effective implementation of subset simulation is considered [1], which relies on the formulation of an equivalent reliability problem [2]. In this setting, failure samples follow the posterior distribution of the original identification problem. Overall, the proposed implementation does not rely on any problem-specific information, avoids the need for knowledge on the maximum likelihood value, and requires minimal modifications to the standard subset simulation algorithm. For an efficient numerical implementation, a parametric reduced-order model formulation based on substructure coupling for dynamic analysis is considered [3]. An application example involving a three-dimensional bridge model equipped with nonlinear devices is presented to illustrate the capabilities of the proposed approach. References: [1] D. J. Jerez, H. A. Jensen, and M. Beer, “An effective implementation of reliability methods for Bayesian model updating of structural dynamic models with multiple uncertain parameters,” Reliability Engineering & System Safety 225:108634, 2022. [2] D. Straub and I. Papaioannou, “Bayesian updating with structural reliability methods,” Journal of Engineering Mechanics 141(3):04014134, 2015. [3] H. A. Jensen and C. Papadimitriou, “Sub-structure coupling for dynamic analysis: Application to complex simulation-based problems involving uncertainty,” Springer-Verlag GmbH, 2019.