Choosing an optimal model for failure data analysis by graphical approach

Abstract

Many models involving combination of multiple Weibull distributions, modification of Weibull distribution or extension of its modified ones, etc. have been developed to model a given set of failure data. The application of these models to modeling a given data set can be based on plotting the data on Weibull probability paper (WPP). Of them, two or more models are appropriate to model one typical shape of the fitting plot, whereas a specific model may be fit for analyzing different shapes of the plots. Hence, a problem arises, that is how to choose an optimal model for a given data set and how to model the data. The motivation of this paper is to address this issue. This paper summarizes the characteristics of Weibull-related models with more than three parameters including sectional models involving two or three Weibull distributions, competing risk model and mixed Weibull model. The models as discussed in this present paper are appropriate to model the data of which the shapes of plots on WPP can be concave, convex, S-shaped or inversely S-shaped. Then, the method for model selection is proposed, which is based on the shapes of the fitting plots. The main procedure for parameter estimation of the models is described accordingly. In addition, the range of data plots on WPP is clearly highlighted from the practical point of view. To note this is important as mathematical analysis of a model with neglecting the applicable range of the model plot will incur discrepancy or big errors in model selection and parameter estimates. © 2013 Elsevier Ltd. All rights reserved. The application of these models to modeling a given data set can be based on plotting the data on Weibull probability paper (WPP). Of them, two or more models are appropriate to model one typical shape of the fitting plot, whereas a specific model may be fit for analyzing different shapes of the plots. Hence, a problem arises, that is how to choose an optimal model for a given data set and how to model the data. The motivation of this paper is to address this issue. This paper summarizes the characteristics of Weibull-related models with more than three parameters including sectional models involving two or three Weibull distributions, competing risk model and mixed Weibull model. The models as discussed in this present paper are appropriate to model the data of which the shapes of plots on WPP can be concave, convex, S-shaped or inversely S-shaped. Then, the method for model selection is proposed, which is based on the shapes of the fitting plots. The main procedure for parameter estimation of the models is described accordingly. In addition, the range of data plots on WPP is clearly highlighted from the practical point of view. To note this is important as mathematical analysis of a model with neglecting the applicable range of the model plot will incur discrepancy or big errors in model selection and parameter estimates.Keywords: Weibull models, Failure data analysis, Model selection, Parameter estimation, Weibull probability paper NOTATION...

An additive modified Weibull distribution

“…The calculation of AIC, BIC or AICc value is based on the estimates of model parameters and sometimes it may be a difficult task. There are also some other methods for final identification [21,30,42,43], but the use of them is either too complicated or very limited. For saving computational cost to obtain the best model, some other effective tools both for reducing the number of initial selected models and saving computational cost of final identification are worth to be explored and we need to know more about the proprieties of the models.…”

“…Lemonte [13] proposed a new three-parameter exponential-type distribution which can have constant, decreasing, increasing, upside-down bathtub and bathtub-shaped hazard rate function. The characteristics of Weibull-related models with more than three parameters were summarized by Zhang and Dwight [30] and a new method for model selection was proposed by graphical approach [30]. Cordeiroa et al [40] introduced the exponential-Weibull lifetime distribution which can only have an increasing or decreasing hazard rate.…”

Investigation of the relationship between reliability of track mechanism and mineral dust content in rocks of lignite open pits

“…According to Kolmogorov-Smirnov statistical test, shows that empirical data to be best modeling with Weibull distribution. In articles [7,29] authors suggests that the Weibull distribution is commonly used to model and analyze the failure data and lifetime data in general.…”

Estimation of the lifetime distribution of mechatronic systems in the presence of a covariate: A comparison among parametric, semiparametric and nonparametric models

“…A discussion about the application of BIC and AIC is described in [30]. Some other model selection procedures were proposed to select between different transformed Weibull models in a non-regression context in [31,32].…”