Graduate Thesis Or Dissertation

 

Machine Learning Applied to Non-periodic Event Detection – A Special Case in Structural Health Monitoring Público Deposited

Contenido Descargable

Descargar PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/44558m46b

Header

Attribute Name LabelAttribute Values Label
Creator
Abstract
  • Structural Health Monitoring (SHM) is known as the process of implementation of damage detection and characterization strategy. The method is widely used in modern critical equipment to minimize the risk of failures. While a significant effort has been devoted by researchers for the detection of faults in situations including periodic dynamic loads, there are no generally recognized methods for the identification of a non-periodic event, especially those with low signal-to-noise ratio (SNR). In this research, the case of an impact on wind turbines blades, as a typical non-periodic event during normal turbine operations, was studied using the newly developed advanced SHM method with implementation of support vector machine, a machine learning algorithm, explicitly, developed for the case with low SNR. Field tests were performed to collect data from vibration sensors installed on blades with artificial impacts obtained by launching tennis balls toward the blades’ trajectory. Pre-processing showed that nearly half of the recorded impact events were successfully identified by visual inspection or by performing short-time Fourier transform. The present research covers visually undetectable impact events, masked under background noise due to low SNR. Numerically simulated impacts on blades at various levels of SNR were used to perform an analysis of various training methods for the machine learning impact detection algorithm. Performance of the trained prediction model was evaluated using filed experimental data. Results on the feasibility and the efficiency of new proposed support vector machine algorithm including its optimization and accuracy were reported.
Contributor
License Label
Resource Type
Fecha de Emisión
Degree Level
Degree Name
Degree Field
Degree Grantors
Graduation Year
Contributor Advisor
Contributor Committeemember
Declaración de derechos
Publisher
Peerreviewed
Language

Relaciones

Relationships Parent Rows Label

Rows Empty Text

En Collection:

Elementos