A novel q-exponential based stress-strength reliability model and applications to fatigue life with extreme values

• Autor(a) principal: SALES FILHO, Romero Luiz Mendonça
• Data de Publicação: 2016
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFPE
• dARK ID: ark:/64986/0013000015r4j
• Texto Completo: https://repositorio.ufpe.br/handle/123456789/17632
• topic: Q-Exponencial; Confiabilidade Força-Estresse; Estimador de Máxima Verossimilhaça; Nelder-Mead; Particle Swarm Optimization; Q-Exponencial; Confiabilidade Força-Estresse; Estimador de Máxima Verossimilhaça; Nelder-Mead; Particle Swarm Optimization

In recent years, a family of probability distributions based on Nonextensive Statistical Mechanics, known as q-distributions, has experienced a surge in terms of applications to several fields of science and engineering. In this work the _-Exponential distribution will be studied in detail. One of the features of this distribution is the capability of modeling data that have a power law behavior, since it has a heavy-tailed probability density function (PDF) for particular values of its parameters. This feature allows us to consider this distribution as a candidate to model data sets with extremely large values (e.g. cycles to failure). Once the analytical expressions for the maximum likelihood estimates (MLE) of _-Exponential are very difficult to be obtained, in this work, we will obtain the MLE for the parameters of the _- Exponential using two different optimization methods: particle swarm optimization (PSO) and Nelder-Mead (NM), which are also coupled with parametric and non-parametric bootstrap methods in order to obtain confidence intervals for these parameters; asymptotic intervals are also derived. Besides, we will make inference about a useful performance metric in system reliability, the called index __(_, where the stress _ and strength are independent q-Exponential random variables with different parameters. In fact, when dealing with practical problems of stress-strength reliability, one can work with fatigue life data and make use of the well-known relation between stress and cycles until failure. For some materials, this kind of data can involve extremely large values and the capability of the q- Exponential distribution to model data with extremely large values makes this distribution a good candidate to adjust stress-strength models. In terms of system reliability, the index _ is considered a topic of great interest, so we will develop the maximum likelihood estimator (MLE) for the index _ and show that this estimator is obtained by a function that depends on the parameters of the distributions for and _. The behavior of the MLE for the index _ is assessed by means of simulated experiments. Moreover, confidence intervals are developed based on parametric and non-parametric bootstrap. As an example of application, we consider two experimental data sets taken from literature: the first is related to the analysis of high cycle fatigue properties of ductile cast iron for wind turbine components, and the second one evaluates the specimen size effects on gigacycle fatigue properties of high-strength steel.