Degradation Analysis for Heterogeneous Data Using Mixture Model

New product testing presents a significant challenge to manufacturers of highly reliable products. For products with high reliability, very few or even no failures might be expected in reliability testing, resulting in limited information about reliability of the products. If a degradation measure that is closely related to failure can be monitored, degradation analysis, as an alternative to the failure time analysis, may lead to improved reliability inference and provide additional information related to the failure mechanisms.Degradation analysis has attracted considerable attention in recent years. Mixed-effects models have been frequently employed to analyze repeated-measures degradation data of multiple units. In existing studies, the test units are usually assumed to be sampled from a homogeneous population, and the random effects in the degradation models are generally assumed to be normally distributed. However, in practical applications of degradation analysis, excessive variability among the degradation paths of different units may be observed due to different reasons such as quality, degradation mechanism, and external environmental condition, etc. The normal distribution may not be adequate to describe the observed unit-to-unit variability. Reliability analysis for units from a nonhomogeneous population with subgroups has been considered only in failure time analysis.This thesis considers the degradation analysis for units coming from a nonhomogeneous population. Instead of the normal distribution, this thesis assumes a normal mixture distribution for the random effects. Both maximum likelihood and Bayesian approaches are adopted in this thesis for the inference of the model parameters and for the derivation of the lifetime distribution. Practical example is used to illustrate the usefulness of the proposed methods. The results show that: (1) both ML method and Bayesian approach give similar estimates for the model parameters; (2) the mixture model has its notable advantages in fitting heterogenous data and providing detailed information regarding different subgroups.

Name thesis:degree_name

Master of Science (MS)

Level thesis:degree_level

masters

Discipline thesis:degree_discipline

Industrial and Systems Engineering (Engineering and Technology)

Grantor dc:publisher

Ohio University

Year dc:date

2013

Author dc:creator

• Ji, Yizhen

Contributors dc:contributor

• Yuan, Tao

• Industrial Engineering
• Bayesian hierarchical model
• Degradation analysis
• Mixture model
• EM algorithm

Statement dc:rights

• unrestricted
• This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws.

Language dc:language

English

Repository record dc:identifier

http://rave.ohiolink.edu/etdc/view?acc_num=ohiou1363310469

OAI identifier oai:identifier

oai:etd.ohiolink.edu:ohiou1363310469