MS20.2 - UQ and Probabilistic Learning in Computational Dynamics

In experimental structural dynamics, reliable estimation of Frequency Response Functions (FRF) is important to correctly characterize a mechanical system. In Experimental Modal Analy-sis (EMA), the FRFs are used as input to a modal parameter estimation algorithm to obtain the modal characteristics of the system. Errors due to noisy measurements are inevitably present in the FRFs and propagate to the modal parameters. A consistent FRF-estimator with low uncer-tainty is therefore needed. Different FRF estimators have been proposed with some consistency when certain noise-related assumptions are fulfilled (H1, H2, etc.). To choose the appropriate frequency response function estimator, some information about the noise in the experimental setup must be known prior to the experiments. In this work we show how to use measurements of ambient noise, made prior to the experiments, to characterize different noise components in the experimental setup. The noise components can be used in the FRF estimation resulting in more accurate estimates. Also, this approach is more general as it relies on fewer noise-related assumptions compared to the conventional estimators. Simulation of an experimental setup based on a simple finite element model is used to present and validate the approach.

Quantification of statistical errors in modal parameter estimates is a topic of intense interest in operational modal analysis, damage diagnosis, modal updating, among many other engineering fields. The uncertainty in the modal parameter estimates stems from the finite data length, unknown, or partly measured inputs, the choice of the identification algorithm, the sensor layout, among other factors. This paper proposes a sensor placement strategy that yields an optimal design of the sensor layout for minimizing the uncertainty when the modal parameters of the structure are obtained from data. The optimization criterion is based on the Fisher information contained in the modal parameter estimates. The related uncertainty is obtained with statistical delta method, which is coupled with subspace identification for modal parameter estimation. The performance of the proposed approach is illustrated on an extensive Monte Carlo simulation of a mechanical system.

Tuned mass dampers are found to be successful in enhancing the structural performance of buildings subjected to wind or seismic loads, but their sensitivity against the uncertainties might strongly reduce their effectiveness. This paper thus provides some insights about the sensitivity of a novel hysteretic tuned mass damper (TMD) in mitigating the seismic response of a prototype building under uncertainties. A two degrees-of-freedom reduced-order model representing a building structure equipped with the TMD is first considered in such a way to facilitate the optimum design of the device’s parameters. Next, a three-dimensional 5-storey laboratory prototype of a steel building equipped with the considered hysteretic TMD is modeled into OpenSees. To this end, a modified Bouc-Wen hysteresis model has been implemented in order to mimic the expected behavior of this novel device. Such numerical model is then employed to investigate the seismic effectiveness of the hysteretic TMD when the parameters of the protected building are uncertain. Several sensitivity analysis techniques are adopted to rank the uncertain parameters of the structure that can affect the performance of the hysteretic TMD. The critical examination of the final results demonstrates that structural damping and floor mass eccentricity along the direction orthogonal to the horizontal seismic ground motion component are the parameters that mostly influence the seismic effectiveness of the considered hysteretic TMD.