多阶段随机退化设备剩余寿命预测方法  

A residual useful life prediction approach for equipments with multi-state stochastic degradation

作　　者：张正新 胡昌华 高迎彬 陈墨 Zhang Zhengxin Hu Changhua Gao Yingbin Chen Mo(Department of Automation Engineering, Rocket Force University of Engineering, Xi＇an 710025, China)

机构地区：[1]火箭军工程大学控制工程系,陕西西安710025

出　　处：《系统工程学报》2017年第1期1-7,共7页Journal of Systems Engineering

基　　金：国家自然科学杰出青年基金资助项目(61025014);国家自然科学基金资助项目(61146030;61074072;61374120)

摘　　要：剩余寿命预测是系统视情维护和健康管理的重要基础.针对设备退化过程中表现出的阶段性差异,提出了一种基于累加和检验变点分析与漂移布朗运动的剩余寿命预测方法.该方法首先对获取的状态检测信息进行累加和检验,获得退化过程的最新显著变点;然后利用最新变点之后的状态检测数据对漂移布朗运动模型的参数进行极大似然估计;最后利用失效阈值的首达时间分布预测设备的剩余寿命.连续搅动水箱式反应堆仿真试验说明了所提方法的有效性.试验结果表明,所提方法能够准确地检测出退化过程的显著变化点,预测设备的剩余寿命,且预测结果波动较小,具有更好的鲁棒性.Residual useful life （RUL） prediction is an important foundation of condition based maintenance and health management of an equipment. Aiming at the multi-state differences during the degradation, a drifted Brownian motion based method incorporating a cumulative sum （CUSUM） test change point analysis was pretended for RUL estimation in this paper. Specifically, a CUSUM test was firstly utilized on the degradation data to get the latest significant change point, based on which parameters of the drifted Brownian motion were estimated through a maximum likelihood estimator. Then, the RUL defined as the first hitting time distribution of a preset threshold was estimated. The proposed method can not only detect the significant change point during degradation precisely, but also prediction the RUL with a lower fluctuation. A continuous stirred-tank reactor （CSTR） simulation study was provided to demonstrate the effectiveness and robustness of the proposed method.

关 键 词：多阶段退化建模 剩余寿命预测 变点分析 状态检测 漂移布朗运动 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]