论文部分内容阅读
针对先验信息为加速退化数据的情况,提出了利用非共轭先验分布进行Bayesian统计推断的剩余寿命预测方法。不预先假定Wiener过程参数值的分布类型,利用加速系数将加速应力下的参数值折算到工作应力水平下,进而使用Anderson-Darling方法确定参数值的最优拟合分布类型。在对参数值进行折算时,根据周源泉提出的理论对Wiener过程参数与加速应力之间的关系进行了推导。参数估计时,通过极大似然法得到超参数的估计值,利用WinBUGS软件实现Markov Chain Monte Carlo仿真得到参数的后验均值。通过某型军用电连接器寿命预测实例验证了所提方法的实用价值和研究意义,结果表明本方法可有效解决先验信息为加速退化数据时进行剩余寿命预测的难题。
Aiming at the fact that prior information is used to accelerate degenerate data, a method of remaining life prediction based on Bayesian statistical inference using non-conjugate prior distribution is proposed. Without presupposing the distribution type of Wiener process parameter values, the accelerating coefficient is used to convert the parameter values under accelerating stress to the working stress level, and then the Anderson-Darling method is used to determine the optimal fitting distribution type of parameter values. Based on the theory proposed by Zhou Yuanquan, the relationship between Wiener process parameters and acceleration stress is derived when the parameter values are converted. When parameters are estimated, the estimation of hyperparameters is obtained by the maximum likelihood method, and the posterior mean of the parameters is obtained by Markov Chain Monte Carlo simulation with WinBUGS software. The practical value and research significance of the proposed method are validated by the life prediction of a certain type of military electrical connector. The results show that this method can effectively solve the problem of predicting the remaining life in order to accelerate degradation data.