【作者】 吕克洪；

【导师】 邱静；

【作者基本信息】 国防科学技术大学 ， 机械电子工程， 2008， 博士

【摘要】 机内测试(Built-in Test, BIT)虚警率高和部件故障预测难一直是故障测试与诊断领域所面临的两个共性问题,阻碍着BIT及故障预测系统效能的充分发挥和更广泛的应用。尤其是近年来,故障预测与健康管理(Prognostics and Health Management, PHM)等新概念的提出对如何降低BIT虚警率、提高部件的故障预测能力提出了更高的要求。时间应力是指设备在生产、运输和工作过程中所承受的各种环境应力和工作应力的时间历程。研究表明：实时应力是导致系统暂时失效进而造成BIT虚警的重要因素,时间应力历程是导致部件性能退化甚至失效的直接外因,对时间应力进行分析可以有效减少BIT虚警和实现部件的故障预测。本文针对目前BIT虚警率高和部件故障预测难的问题,系统地分析时间应力诱发故障和BIT虚警的机理与规律,对基于时间应力分析的BIT降虚警技术和故障预测技术进行深入研究。论文的主要研究内容包括：1.时间应力诱发故障和BIT虚警的机理分析与建模系统地分析了时间应力诱发故障和BIT虚警的机理和一般规律；建立了基于支持向量机(Support Vector Machine, SVM)的时间应力与BIT虚警的关联模型,描述了时间应力大小与BIT虚警的关联关系。建立了基于随机Petri网的时间应力与损伤关系的动态描述模型,获得时间应力历程与损伤问的转化关系。建立了基于损伤的故障演化隐马尔可夫模型(Hidden Markov Model, HMM),描述了损伤与故障间的演化关系。2.基于时间应力分析的BIT降虚警技术研究(1)提出了基于故障模式、影响及应力分析(Failure Mode, Effects and Stress Analysis, FMESA)和多元Logistic回归的关联分析方法,从定性和定量两个角度分析导致BIT虚警的主要时间应力因素和相应的影响程度。针对时间应力导致BIT虚警的主要应力区域确定的问题,提出了基于SVM的关联阈值优化选取方法,获得了时间应力导致BIT虚警的时间应力上下限阈值。(2)提出了基于时间应力分析的SVM虚警识别方法,能有效识别关联区域内由时间应力造成的虚警。由于该方法不能有效识别关联区域内的真实故障和关联区域外的虚警,分别提出了基于时间应力分析的SVM-HMM虚警识别方法和基于核主元聚类分析(Kernel Based Principal Component Clustering Analysis, KPCCA)模型和HMM串联的虚警识别方法。最后,综合上述三种情形,提出了基于时间应力分析的SVM-KPCCA-HMM综合虚警识别方法,可在时间应力的全区域内有效识别BIT虚警和真实故障。3.基于时间应力分析的故障预测技术研究(1)针对如何将部件所承受的应力历程转化为损伤的问题,以被监控对象的失效物理模型为基础,提出了基于多组件寿命消耗监控(Multi-component Life Consumption Monitoring, MLCM)的累积损伤信息提取方法,并以典型的连接组件和分离组件为对象对该方法的技术步骤进行了具体研究,结果表明,该方法可有效地将部件所承受的应力历程转化为部件的累积损伤度量。(2)综合考虑历史变化趋势和应力突变的影响,提出了基于多点损伤动态优化自回归(Auto-Regression, AR)模型的故障预测方法。由于累积损伤分析和计算等过程中存在不确定性和随机性,为进一步提高组件故障预测的准确性,提出了基于优化AR-HMM的故障预测方法。验证结果表明,该方法具有较高的预测置信度。4.基于时间应力分析的BIT降虚警和故障预测技术的应用与验证以某直升机航向姿态系统为对象,设计并实现了具有BIT虚警识别能力和关键部件故障预测能力的PHM系统,并通过理论分析和试验对本文所研究的虚警识别方法和故障预测方法进行了验证。结果表明,本文所研究方法具有较好的虚警识别和故障预测效果。更多还原

【Abstract】 The high False Alarm Rate (FAR) in Built-in Test (BIT) and low fault prognostics ability of equipments are two common key problems in the fault diagnosis field. They directly prevent BIT and fault prognostics system from more extensive application. In recent years, with some new conceptions such as Prognostics and Health Management (PHM) appearing, how to reduce FAR of BIT and improve the fault prognostics ability become more and more important.The time stress include all kinds of environment and operating stress such as shock, vibration, temperature and so on that the equipment suffered in the manufacture, transport and operating process. On the one hand, the real-time time stress is the main factor that disables system temporarily and incurs the BIT false alarms. On the other hand, the time stress directly induce degradation or deviation of the components from an expected normal condition. Therefore, analyzing the time stress can reduce false alarms of the BIT and improve fault prognostics ability effectively.This paper is aiming to solve the two problems of BIT’s high FAR and equipment’s low fault prognostics ability in fault diagnosis field. The mechanisms of the faults and BIT false alarms induced by time stress are analyzed systematically. Then, the BIT false alarm reducing and fault prognostics technologies based on time stress analysis are studied.The main contents of the dissertation are as follows.1. The mechanisms of how the time stress induces faults and BIT false alarms are analyzed systematically and some relative models are built.Firstly, the mechanisms of the faults and BIT false alarms induced by time stress are analyzed in detail. In order to analyze the relationship between time stress and BIT fault alarm, a relevance analysis model based on Support Vector Machine (SVM) is built. The model can acquire the main time stress region that induces the BIT false alarms. Secondly, a dynamic description damage model is researched based on Generalized Stochastic Petri Nets (GSPN). This model can analyze the evolution relationship between time stress and component’s damage. Finally, a probability fault evolution model based on Hidden Markov Model (HMM) is studied. This model can analyze the evolution process of component’s fault.2. The BIT false alarm reducing technology based on time stress analysis is studied.(1) Two relevance analysis methods between time stress and BIT false alarm based on Failure Mode, Effect and Stress Analysis (FMESA) and logistic regression are studied. These two methods can analyze the main stress factor that induce BIT false alarm from qualitative and quantitative aspects. In order to acquire the optimization threshold values of the relevance analysis model between time stress and BIT false alarm, the threshold values optimization selection method is discussed in detail.(2) A false alarm recognition method based on SVM and time stress analysis is studied, and it can recognise the false alarms in the time stress relevance region. But this method can not recognise the real faults in the time stress relevance region and the false alarms out of the time stress relevance region effectively. In order to solve the first problem, a false alarm recognition method based on SVM-HMM and time stress analysis is proposed. In order to solve the second problem, a false alarm recognition method based on Kernel based Principal Component Clustering Analysis (KPCCA) and HMM is studied. Finally, a synthetical false alarm recognition method based on SVM-KPCCA-HMM and time stress analysis is proposed. And it can recognise the false alarms and real faults in the whole time stress region.3. The fault prognostics technology based on time stress analysis is studied.(1) An accumulative damage assessment method based on component’s failure models and Multi-component Life Consumption Monitoring (MLCM) is studied. Then, the accumulative damage assessment methods of joint and separate components are studied in detail. It can transform the time stress suffered in the life time into equipment’s damage.(2) Considering the history current of damage and the abrupt change affection of time stress, a fault prognostics method based on dynamic optimization auto-regression (AR) model and multipoint damages is studied. Because there are some uncertainties in the damage assessment process, a fault prognostics method based on optimization AR model and HMM is proposed. This method can improve the accuracy. The experimental results show that this method has a higher degree of confidence.4. The BIT false alarm reducing and fault prognostics technologies based on time stress are applied and validated.In order to evaluate the efficiency of the BIT false alarm reducing and fault prognostics technologies, a PHM system for heading attitude system of the helicopter is built and implemented. Then, some time stress experimental studies on the heading attitude system are executed. The results show that BIT false alarm reducing technology in this paper can reduce false alarms of BIT effectively, and the fault prognostics technology has a good fault prognostics effect.更多还原