Degree Type

Dissertation

Date of Award

2019

Degree Name

Doctor of Philosophy

Department

Statistics

Major

Statistics; Wind Energy Science, Engineering, and Policy

First Advisor

William Q. Meeker

Abstract

Many modern engineering systems are being evaluated with prognostics and health management (PHM) tools. The goal of PHM is to evaluate and predict the state of a system or product during its service life. PHM aims to predict failure in order to alleviate system risks. Many different industries, including wind energy use PHM systems. In wind power generation, PHM aims to reduce maintenance costs and extend the availability periods for each turbine. With large technological advancements, the upcoming generation of reliability field data will be abundant with valuable information to help predict future failures of many systems including wind turbines. Sensors and tools such as synchrophasors (PMUs), accelerometers, programmable logical controllers (PLCs), and so on help monitor variables such as vibration, use rate, environmental variables, and much more. Such sensors and tools provide information about component or system use, load, and operating environment over time. Such multivariate time series data are referred to as system operating/environmental data (SOE data). Wind turbines collect these data through supervisory control acquisition data (SCADA) systems. Such data, along with failure data can be used to build models that can be used to predict the remaining life of individual wind turbines. Throughout the model building process: statistical, probabilistic, and model uncertainties should be assessed to properly determine the accuracy of all predictions calculated. In this thesis, three real application problems are presented. In Chapter 2, we focus on reliability small wind turbine (SWT) reliability issues and look at recurrence data from 21 individual 100kW wind turbines. We outline a nonhomogenous Poisson process (NHPP) model with a Bayesian hierarchical power law structure in discrete-time that allows for inclusion of time-varying covariates. Data used in Chapter 2 was provided by a power systems company in the United States. In Chapter 3, we focus on developing a repairable systems simulation tool in R to assist in maintenance decisions for wind turbine owners and operators. In Chapter 4 we introduce a nonparametric algorithm for predicting future part consumption in systems that fail from multiple failure types. Chapter 5 summarizes the work in this thesis and discusses the future of these research areas.

Copyright Owner

Michael Stanley Czahor

Language

en

File Format

application/pdf

File Size

125 pages

Share

COinS