TY - THES
AU - Attanayake, Dona Nayomi Sandarekha
Y2 - 2020/08/04
Y1 - 2019
UR - http://hdl.handle.net/11343/241687
AB - Operation and maintenance of a fleet always require a high level of readiness, reduced cost, and improved safety. In order to achieve these goals, it is essential to develop and determine an appropriate maintenance programme for the components in use. A failure analysis involving failure model selection, robust parameter estimation, probabilistic decision making, and assessing the cost-effectiveness of the decisions are the key to the selection of a proper maintenance programme. Two significant challenges faced in failure analysis studies are, minimizing the uncertainty associated with model selection and making strategic decisions based on few observed failures. In this thesis, we try to resolve some of these problems and evaluate the cost-effectiveness of the selections. We focus on choosing the best model from a model space and robust estimation of quantiles leading to the selection of optimal repair and replacement time of units. We first explore the repair and replacement cost of a unit in a system. We design a simulation study to assess the performance of the parameter estimation methods, maximum likelihood estimation (MLE), and median rank regression method (MRR) in estimating quantiles of the Weibull distribution. Then, we compare the models; Weibull, gamma, log-normal, log-logistic, and inverse-Gaussian in failure analysis. With an example, we show that the Weibull and the gamma distributions provide competing fits to the failure data. Next, we demonstrate the use of Bayesian model averaging in accounting for that model uncertainty. We derive an average model for the failure observations with respective posterior model probabilities. Then, we illustrate the cost-effectiveness of the selected model by comparing the distribution of the total replacement and repair cost. In the second part of the thesis, we discuss the prior information. Initially, we assume, the parameters of the Weibull distribution are dependent by a function of the form rho = sigma/mu and re-parameterize the Weibull distribution. Then we propose a new Jeffreysâ€™ prior for the parameters mu and rho. Finally, we designed a simulation study to assess the performance of the new Jeffreysâ€™ prior compared to the MLE.
KW - Maximum likelihood estimation
KW - Bayesian Analysis
KW - Bayesian Model Averaging
KW - Median Rank Regression
KW - Failure Analysis
KW - Weibull distribution
KW - Log-location scale distributions
KW - Jeffreys' Prior
KW - Cost Analysis
KW - Robustness
KW - Parameter estimation
KW - Probabilistic decision making
KW - Censored data
T1 - Risk Analysis and Probabilistic Decision Making for Censored Failure Data
L1 - /bitstream/handle/11343/241687/204beeaf-4faf-e911-94a8-0050568d7800_Risk%20Analysis%20and%20Probabilistic%20Decision%20making%20for%20Censored%20Failure%20data.pdf?sequence=1&isAllowed=y
ER -