A Tukey's g-and-h distribution based approach with PSO for degradation reliability modeling. Issue 5 (7th August 2019)

• Record Type: Journal Article
• Authors: Zhang, Jinbao; Zhao, Yongqiang; Liu, Ming; Kong, Lingxian
• Is Part Of: Engineering computations. Volume 36:Issue 5(2019)
• Page Start: 1699
• Page End: 1715
• Publication Date: 2019-08-07
• Subjects: PSO -- Quantile regression -- Degradation path -- Reliability estimation -- Tukey's g-and-h distribution; Computer-aided engineering -- Periodicals; Computer graphics -- Periodicals; 620.00285
• DOI: 10.1108/EC-11-2017-0428
• Languages: English
• ISSNs: 0264-4401

Abstract

Abstract : Purpose: A generalized distribution with wide range of skewness and elongation will be suitable for the data mining and compatible for the misspecification of the distribution. Hence, the purpose of this paper is to present a distribution-based approach for estimating degradation reliability considering these conditions. Design/methodology/approach: Tukey's g-and-h distribution with the quantile expression is introduced to fit the degradation paths of the population over time. The Newton–Raphson algorithm is used to approximately evaluate the reliability. Simulation verification for parameter estimation with particle swarm optimization (PSO) is carried out. The effectiveness and validity of the proposed approach for degradation reliability is verified by the two-stage verification and the comparison with others' work. Findings: Simulation studies have proved the effectiveness of PSO in the parameter estimation. Two degradation datasets of GaAs laser devices and crack growth are performed by the proposed approach. The results show that it can well match the initial failure time and be more compatible than the normal distribution and the Weibull distribution. Originality/value: Tukey's g-and-h distribution is first proposed to investigate the influence of the tail and the skewness on the degradation reliability. In addition, the parameters of the Tukey's g-and-h distribution is estimated by PSO with root-mean-square error as the object function.