考虑随机退化和信息融合的剩余寿命预测方法  

Residual Lifetime Prediction Method with Random Degradation and Information Fusion

作　　者：蔡忠义[1] 陈云翔[1] 李韶亮 项华春[1] 王泽洲[1] 

机构地区：[1]空军工程大学装备管理与安全工程学院,西安710051 [2]空军驻北京地区军事代表室,北京100024

出　　处：《上海交通大学学报》2016年第11期1778-1783,共6页Journal of Shanghai Jiaotong University

基　　金：国防预研项目(51327020104)

摘　　要：针对性能退化过程服从Wiener过程的产品,运用贝叶斯统计推断法,提出了一种融合产品现场实测性能退化数据与同类产品常规退化试验信息、历史寿命信息的个体剩余寿命预测方法.建立了基于Wiener过程的产品剩余寿命模型;考虑到个体之间的性能退化差异性,假定Wiener过程参数服从随机分布模型,建立了个体现场实测退化数据下分布参数的贝叶斯估计模型,给出了超参数后验估计公式;分别建立了退化数据和寿命数据下的完全似然函数,构建了基于最大期望算法的超参数先验估计模型;通过实例分析验证了所提方法的正确性和优势,结果表明本方法可有效处理个体现场实测退化信息与同类产品先验信息之间的剩余寿命预测问题.Aim at the product whose its performance degradation process obeys the Wiener process,the individual residual lifetime prediction method for field measurment of performance degradation data of a product,the normal degradation test data and the history lifetime data of like products was proposed.The residual lifetime model of the product was built based on the Wiener process.Considering the performance degradation variance among individuals,the parameters of Wiener process was assumed to be random distribution model.A Bayesian estimation model of distribution parameters was built under individual field measured degradation data.The posterior estimation formula of hyper parameter was obtained.The entire likelihood functions of degradation data and lifetime data were established.A prior estimation model of hyper parameter was also built based on the expectation maximization（EM）algorithm.The accuracy and superiority of the present method was verified by an example.The result shows that this method can deal with the residual lifetime prediction problem for field measured degradation data of individual and prior data of like products.

关 键 词：剩余寿命预测 贝叶斯推断 性能退化数据 历史寿命数据 最大期望算法 退化差异 

分 类 号：TB114.3[理学—概率论与数理统计]