Tuesday  |  Salon 9  |  10:20 AM–10:40 AM

#17098–Model Class and Parameter Selection for Bayesian Filtering with Application to a Modular Active Spring-Damper System: Round-Robin Challenge

Bayesian filtering techniques recursively combine a mathematical model and a system's response measurements to improve the model's estimation and prediction capabilities. These techniques are used for applications such as structural monitoring, control, and condition assessment of civil and mechanical structures subjected to dynamic excitations. Filtering is required because it is generally not feasible or practical to directly measure the complete response and model parameters of a system of interest due to economic or physical constraints. Instead, noise-contaminated response measurements at limited spatial locations are available, and the objective is to extract information about the system's response and model parameters. To achieve this, a reduced-order surrogate model must be used, which imposes a constraint on the class of models used for estimation. Choosing a model class and its parameters with enough complexity is necessary to select an appropriate surrogate model to represent the physical system. However, increasing the number of model parameters reduces computational efficiency and makes modeling and filtering more challenging. To address this challenge, this paper presents the application of Bayesian filtering for parameter estimation and model uncertainty characterization of a large-scale suspension strut system called the Modular Active Spring-Damper System (MASDS). The MASDS (which is developed in academic settings to study the uncertainty in a suspension strut's dynamic outputs) is similar to an aircraft landing gear and includes an upper space truss structure with added payload, a suspension system with stiffness and damping components, a lower space truss structure, guidance links for kinematic motion, an elastic foot with stiffness and damping components, and a user-specified drop height. This paper compares the performance of various Bayesian filters in terms of computational efficiency, estimation accuracy, and statistical error behavior in the identification of the MASDS, using models at multiple fidelity levels.

Aleem Ullah   University of Nebraska–Lincoln
Milad Roohi   University of Nebraska–Lincoln