Publications

Selecting and Identifying Hysteretic Models for Steel Damper Pairs Using Full-Scale Testing Data

Hysteresis is an intrinsic phenomenon for many nonlinear base isolators. Since numerous models with different strengths and weaknesses have been proposed to describe hysteresis, selecting the most appropriate model, a trade-off between model complexity and accuracy, becomes an important task. This study performs a comprehensive Bayesian model selection on restoring force-displacement data from two steel damper pairs utilized in a hybrid base-isolation system during full-scale shake table testing. To ensure a breadth of models are examined, six models are considered, including equivalent linear models, models from the Bouc–Wen family, and the recently proposed Vaiana–Rosati model. A nested sampling algorithm is used to calculate evidence values for each model under four different ground motions at two different levels of intensity. The results show that the Bouc–Wen family of models consistently yields the highest posterior model probabilities using both Bayesian model selection as well as other model selection criteria. Despite their added parameterizations, the degrading and the generalized Bouc–Wen models boast greater plausibility than the conventional Bouc–Wen model due to their ability to better capture the energy dissipation and asymmetrical hysteresis. Model preference is partially based on intensity, with more intense tests selecting the degrading Bouc–Wen model while lower intensity tests choose the generalized Bouc–Wen, but no clear threshold or delineation is found. Further, model preference is not found to be closely associated with a measure of dissipated hysteretic energy.

Hierarchical Bayesian calibration of Bouc–Wen hysteretic models with applications to seismic isolators

Harnessing the potential of base-isolation devices requires both experimental testing campaigns to observe their behavior as well as properly calibrated hysteretic models to predict their response for future hazards. However, calibrating a nonlinear hysteretic model based on a single experimental test might omit critical behaviors observed in other tests. This study performs hierarchical Bayesian calibrations for a series of bi-axial hysteretic models from the Bouc–Wen family that attempt to capture the restoring force behavior observed in two steel yielding damper pairs during a full-scale testing campaign at E-Defense. Critically, the hierarchical approaches incorporate data from numerous tests featuring different ground motions into calibration. Two different treatments of the prediction error variance are explored to cover both non-informative assumptions as well as probabilistic models for the error variance. The results demonstrate that the two different approaches yield fairly comparable measures of parametric uncertainty, e.g., posterior distributions, for a given device, but the different devices actually produce different degrees of uncertainty in the hysteretic shape parameters. Further, the impact of the two different approaches is minimal in terms of the prediction error variance for observed data, but the probabilistic model is much better suited to unobserved data. A joint calibration considering data from both devices also reveals that adding data from a second device does not necessarily result in a reduction in the overall uncertainty of model parameters or response predictions.

Modal analysis of Doppler shift and strain time histories extracted from distributed fiber optic sensing data

While fiber optic sensing systems (FOSS) operate through a variety of technologies, optical frequency domain reflectometry (OFDR) has emerged as a means for distributed sensing, wherein a single fiber contains hundreds to thousands of sensing points along its length. Currently, most OFDR distributed sensing focuses on strain sensing; however, prior to deriving strain information, the changes in the carrier frequency of optical gratings can also be monitored. These changes in carrier frequency register as Doppler shifts, and this study explores treating these shifts like dynamic signals, e.g. time histories. Further, this study demonstrates that the Doppler shift time histories contain modal information distinct from strain measurements that can be similarly extracted through traditional modal analysis. Using experimental data from modal testing of a cantilevered plate, frequency-based modal identification is performed on both strain and Doppler shift time histories recovered from an OFDR-based distributed FOSS. The results reveal that Doppler shift time histories yield estimates of the slope mode shapes, as opposed to the strain mode shapes produced by the strain time histories, which can be used to recover displacement mode shapes through numerical integration. In addition, comparisons between the dynamic signals show that the Doppler shift time histories provide a greater number of identifiable modes with higher correlations to reference mode shapes than strain measurements.

A bi-fidelity DeepONet approach for modeling hysteretic systems under uncertainty

This study proposes a multi-fidelity paradigm for developing surrogates of degrading hysteretic systems under uncertainty through the use of deep operator networks (DeepONets). Instead of attempting to directly train a DeepONet on the original response, this study adopts a residual modeling approach wherein the DeepONet is trained on the discrepancy between the original (high-fidelity) data source and a relatively simpler (low-fidelity) representation of the system. Within these examples, a conventional Bouc-Wen model is treated as a “low-fidelity” representation given that it is free of any further assumptions about the nonlinear behavior, while the “high-fidelity” data is generated from different structures with various forms of complex hysteretic behavior. The results of this study show that the proposed multi-fidelity approach consistently outperforms standard surrogates trained on only the original datasets considering a variety of systems with unknown parameters. The results also show that the difference in performance grows as training data becomes more scarce, a critical consideration for many real-world engineering systems, and that the proposed multi-fidelity approach maintains its performance edge even when controlling for training time and noise in the training data.

Assessing the reliability of the cargo deck of a hovercraft through fiber optic strain measurements

While traditional sensing systems can suffer from low spatial sampling density and often face significant operational challenges in the maritime environments routinely experienced by naval craft, fiber optics provide a distributed sensing solution that is insensitive to many of these environmental stressors. Through the use of fiber optic sensing systems, strain data was collected along the cargo deck of two naval hovercraft during various maneuvers and loading conditions. This work presents a methodology for translating those strain measurements into deflection estimates, and ultimately reliability analyses, through optimization and analytical modeling tools. Specifically, by treating the cargo deck of the hovercraft as a large, thin plate, direct relationships between strain, bending curvature, and deflection can be reliably established. Comparisons of probability distributions of maximum absolute deflections experienced during set maneuvers and loading conditions reveal differences in the response between the two craft. These differences are further explored in the context of reliability indices based on limit deflections.

Simulating pitting corrosion in AM 316L microstructures through phase field methods and computational modeling

This work investigates how the crystallographic features of additive manufactured (AM) microstructures impact the pitting corrosion process through computational simulations of phase field models. Crystallographic influence is explored by introducing orientation dependencies into the corrosion potentials and elastic constants of the model through microstructural data provided from AM 316L samples. Comparisons of evolved pit morphologies and stress responses are made to a standard homogeneous, semi-circular model to better highlight how the complexity of AM microstructures affects pit evolution and stress concentrations. The results illustrate that AM-informed modeling cases produce larger, deeper pits with numerous locations of elevated stress concentrations along the pit front.