Bootstrap analysis of designed experiments for reliability improvement with a non-constant scale parameter. (April 2017)

• Record Type: Journal Article
• Authors: Wang, Guodong; He, Zhen; Xue, Li; Cui, Qingan; Lv, Shanshan; Zhou, Panpan
• Is Part Of: Reliability engineering & system safety. Volume 160(2017)
• Page Start: 114
• Page End: 121
• Publication Date: 2017-04
• Subjects: Design of experiments -- Censored data -- Percentile -- Weibull distribution; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2016.12.006
• Languages: English
• ISSNs: 0951-8320

Abstract

Abstract: Factors which significantly affect product reliability are of great interest to reliability practitioners. This paper proposes a bootstrap-based methodology for identifying significant factors when both location and scale parameters of the smallest extreme value distribution vary over experimental factors. An industrial thermostat experiment is presented, analyzed, and discussed as an illustrative example. The analysis results show that 1) the misspecification of a constant scale parameter may lead to misidentify spurious effects; 2) the important factors identified by different bootstrap methods (i.e., percentile bootstrapping, bias-corrected percentile bootstrapping, and bias-corrected and accelerated percentile bootstrapping) are different; 3) the number of factors affecting 10th percentile lifetime significantly is less than the number of important factors identified at 63.21th percentile. Highlights: Product reliability is improved by design of experiments under both scale and location parameters of smallest extreme value distribution vary with experimental factors. A bootstrap-based methodology is proposed to identify important factors which affect 100 p th lifetime percentile significantly. Bootstrapping confidence intervals associating experimental factors are obtained by using three bootstrap methods (i.e., percentile bootstrapping, bias-corrected percentile bootstrapping, and bias-corrected and accelerated percentile bootstrapping). The important factors identified by different bootstrap methods are different. The number of factors affecting 10th percentile significantly is less than the number of important factors identified at 63.21th percentile.