Combining generalized renewal processes with non-extensive entropy-based q-distributions for reliability applications
Author
dc.contributor.author
Lins, Isis Didier
Author
dc.contributor.author
Moura, Marcio das Chagas
Author
dc.contributor.author
López Droguett, Enrique
Author
dc.contributor.author
Correa, Thais Lima
Admission date
dc.date.accessioned
2018-10-29T13:57:55Z
Available date
dc.date.available
2018-10-29T13:57:55Z
Publication date
dc.date.issued
2018-04
Cita de ítem
dc.identifier.citation
Entropy Volumen: 20 Número: 4 Número de artículo: 223
es_ES
Identifier
dc.identifier.other
10.3390/e20040223
Identifier
dc.identifier.uri
https://repositorio.uchile.cl/handle/2250/152271
Abstract
dc.description.abstract
The Generalized Renewal Process (GRP) is a probabilistic model for repairable systems that can represent the usual states of a system after a repair: as new, as old, or in a condition between new and old. It is often coupled with the Weibull distribution, widely used in the reliability context. In this paper, we develop novel GRP models based on probability distributions that stem from the Tsallis' non-extensive entropy, namely the q-Exponential and the q-Weibull distributions. The q-Exponential and Weibull distributions can model decreasing, constant or increasing failure intensity functions. However, the power law behavior of the q-Exponential probability density function for specific parameter values is an advantage over the Weibull distribution when adjusting data containing extreme values. The q-Weibull probability distribution, in turn, can also fit data with bathtub-shaped or unimodal failure intensities in addition to the behaviors already mentioned. Therefore, the q-Exponential-GRP is an alternative for the Weibull-GRP model and the q-Weibull-GRP generalizes both. The method of maximum likelihood is used for their parameters' estimation by means of a particle swarm optimization algorithm, and Monte Carlo simulations are performed for the sake of validation. The proposed models and algorithms are applied to examples involving reliability-related data of complex systems and the obtained results suggest GRP plus q-distributions are promising techniques for the analyses of repairable systems.