Date of Award
Spring 2025
Document Type
Open Access Thesis
Department
Nuclear Engineering
First Advisor
Austin Downey
Abstract
Dynamic forces and evolving structural boundary conditions pose challenges for various structural systems such as aircraft, orbital infrastructure, and energy harvesting devices. The design, evaluation, and functionality, of such systems can be aided through the collection and analysis of data. However, real-time decision-making for systems experiencing high-rate changes can pose unique challenges, if assessments are to be made accurately and rapidly enough to be relevant. In cases where the systems are well-defined and thoroughly understood, monitoring the frequency response can be instrumental in determining the state of structures subjected to high-rate structural boundary condition changes. This study focuses on investigating frequency detection methods to enable real-time state estimation for such structures. The research explores progress and findings related to extracting the real-time frequency response of structures. This study compares a novel technique; Delayed Comparison Error Minimization, a more traditional FFT based method, a trained neural network-based method, and a method combining aspects from the Delayed Comparison Error Minimization and neural network techniques in an at-tempt to potentially leverage the strengths of each. The performance of each method will be demonstrated, and the results examined and discussed in terms of latency, precision, and possible applicability. Training of systems that require it will be performed using synthetic data, and the performance of each method demonstrated on a synthetic data set. Each method’s performance will also be evaluated on data collected from the DROPBEAR testbed in order to validate performance on a physical system with a changing boundary condition. The DROPBEAR testbed consists of an oscillating beam with one fixed end and a roller support that moves in a controlled manner along the beam’s length, altering the frequency response of the beam proportionally to the roller location.
Rights
© 2025, James Scheppegrell
Recommended Citation
Scheppegrell, J.(2025). Hybrid Machine Learning and Comparison Error Minimization for Frequency Domain-Based Rapid State Estimation in Structures Subjected to High-Rate Boundary Change. (Master's thesis). Retrieved from https://scholarcommons.sc.edu/etd/8248