基于统计数据驱动的复杂性能退化过程剩余寿命预测方法

Remaining useful lifetime prediction is one key to efficient predictive maintenance. The complexities of degradation processes include the nonlinear, non-Gaussian, and multistage properties; the relationships between multiple degradation processes; and the effect on degradation from external impact. Therefore, more effective remaining useful lifetime prediction methods are needed. Based on the analysis for statistical degradation data, this project will establish nonlinear and non-Gaussian models that can be used to predict the remaining useful lifetime accurately. Adaptive strategy will be adopted to process the multistage property. The models of the relationships between different degradation processes will be studied, and these models will further be used to analyze the competition between different failure modes, as well as to predict the remaining useful lifetime. The hybrid models including discrete impact and continuous degradation processes will also be considered. Based on the hybrid models, Bayesian filtering will be designed to estimate the degradation levels and the distribution of the remaining useful lifetime. After completing this project, we hope that some good results can be obtained. First, we could get some accurate and flexible models that can characterize the nonlinear, non-Gaussian, and multistage properties. Based on these models, we could also provide accurate distributions of the remaining useful lifetime. Second, we could discover how different degradation processes are related, and how impact processes affect degradation processes. Then the effective degradation models and remaining useful lifetime prediction methods could be proposed.

剩余寿命预测技术是实现设备高效预测维护策略的核心技术。非线性、非高斯性、多阶段性等性能退化特征，多性能退化过程相互关联，存在外部冲击过程作用，构成了性能退化过程的复杂性，并对剩余寿命预测技术提出了更高的要求。本项目在分析统计退化数据的基础上，重点研究基于精确剩余寿命分布计算的退化过程非线性、非高斯性特征建模；处理退化过程多阶段性特征的自适应剩余寿命预测方法；多性能退化过程的关联性分析与建模、竞争失效分析及剩余寿命预测方法；离散冲击过程和连续性能退化过程的混合建模、估计离散冲击损伤量和连续退化量的混合贝叶斯滤波算法及混合剩余寿命分布模型。通过本项目的研究，期望能够建立灵活而精确的非线性、非高斯、多阶段性等复杂性能退化特征的模型及相应的剩余寿命分布模型；揭示多性能退化过程相互关联、以及外部冲击过程作用于性能退化过程等复杂现象的机理，并提出相应的性能退化模型和剩余寿命预测方法。

复杂工程系统的寿命预测，是保障工业生产过程安全性的核心技术，在得到了学术界和工业界的广泛关注。本项目围绕“基于统计数据驱动的复杂性能退化过程剩余寿命预测方法”，重点研究了微小故障检测与辨识、基于自适应维纳过程的剩余寿命预测理论等复杂性能退化过程剩余寿命预测理论；同时，将上述理论分别应用于感应电机气隙偏心故障诊断、IGBT性能退化故障诊断、高速列车性能退化故障诊断、大型提升机综合性能故障诊断、大型光伏系统面板清洁度评价、大型汽轮机跳机故障预测等工程问题。项目研究取得了数据驱动的早期微小故障诊断方法、基于随机过程理论的自适应剩余寿命预测方法等理论成果；同时将理论成果同具体的工程应用对象相结合，提出了一系列针对具体工程系统对象的故障诊断与预测技术。相关理论研究成果获得了2018年度教育部自然科学一等奖、2016年度中国自动化学会自然科学一等奖，发表/录用SCI论文6篇。本项目部分研究成果已应用于大型提升机综合性能故障诊断、大型光伏系统面板清洁度评价等工程实际问题，获得中国煤炭工业协会科技进步三等奖，授权发明专利5项，并产生相应的经济效益。经合作企业统计，大型提升机综合性能故障诊断每年能够节约运维费用640万元左右；大型光伏系统面板清洁度评价系统已产生210万元销售额。截止目前为止，项目成果转化为合作企业创造效益800余万元。