设备冷却水泵的故障率时变性研究

设备冷却水泵具有高可靠性、长寿命的特点,历史失效数据的小子样问题突出,并且由冲击、振动、磨损、腐蚀等耗损性因素引起的故障,其故障率具有时变性,随时间的推移而增加.故障率λ为常数的Jeffreys先验模型不能以合理概率复现设备冷却水泵的故障数据,无法满足分析λ时变性的要求.本文在对数线性模型基础上为λ引入额外时变性,建立Poisson分布的广义线性模型,研究λ的时间趋势;不仅利用定性图检验评价模型复现设备冷却水泵故障数据的能力,而且还应用贝叶斯χ~2统计量进行定量化检验;经过定性和定量的双重检验,表明该模型据有良好的预计能力,可以分析设备冷却水泵故障率λ的时变性.

Research of time-varying characteristics for component cooling pump failure rate

Component cooling pump has the characteristics of high reliability and long lifetime, its failure data have obvious small sample problem. The component cooling pump failure are caused by lossy factors of impact, shaking, abrading, corrosion, and the failure rate λ has time trend, the λ will raise (decrease) with the increase of time. The Jeffreys prior model of invariable λ couldn't replicate failure data for component cooling pump very well, so it couldn't analyse the time-varying characteristics of λ. This paper introduced additional time-varying characteristics for λ on the base of logarithmic linear model (LLM), built generalized linear model (GLM) for Poisson distribution and assessed the time trend of λ and inspected the model through posterior predictive distribution, assessed the model ability of replicating observed data by graph inspection and Bayesian chi-square. The results shown GLM had well fit index and was more propitious to assess the failure rate λ of component cooling pump.