复杂装备故障预测与健康管理关键技术研究

发布时间：2019-05-29 22:21

【摘要】：近年来航空、航天、船舶等复杂装备系统的安全性、可靠性成为人们关注的焦点。由于其结构复杂，一旦发生故障则会造成巨大的损失。因此，迫切需要提高复杂装备系统的可靠性、可修复性和安全性。然而，目前故障诊断领域的研究工作主要集中在研究系统的状态评价和故障诊断方面，所关心的是系统“当前”的运行状态，对系统故障预测与健康管理的研究则较少。这种传统的“事后维修”和“计划维修”方法，在应对瞬息万变的新情况时存在严重不足。“视情维修”和“预知维修”将故障消灭在萌芽状态，成为未来系统维护保障的发展方向。所以，近些年关于故障预测与健康管理（Prognostic and Health Management，PHM）方面的研究成为国内外学者的研究热点，PHM技术代表着故障诊断技术从传统的基于传感器的诊断向基于智能系统的预测转变。但是，目前国内在这方面的研究刚刚起步，研究成果较少。本文针对复杂装备PHM技术中涉及的关键理论和技术开展研究，主要内容包括：第一，复杂装备PHM服务模式研究。针对目前故障诊断资源缺乏有效的组织管理，资源共享率低等问题，通过分析复杂装备的特点与传统故障诊断模式的不足，提出一种面向服务的PHM系统架构。研究了系统服务模式和特点，分析了系统构建过程中所需的关键技术及主要功能模块，为复杂装备领域开展公共PHM系统服务平台研究提供一种新思路。第二，数据降维与故障诊断技术研究。故障诊断是PHM的核心内容，针对复杂装备高维、非线性、早期故障难以识别等特点，提出一种基于非线性流形学习和混合隐马尔科夫模型（Hidden Markov Model，HMM）的故障诊断方法。混合HMM模型将系统状态分为正常状态、间歇故障、中间状态和故障状态，以全面反映系统真实状态。通过局部保持投影算法（Locality preserving projection，LPP）将原始高维故障数据映射至低维空间，提取数据的内在流形特征作为特征矢量；利用混合HMM作为分类器实现对各状态的分类识别。通过仿真分析，将该方法与其它方法进行对比，结果表明，LPP-HMM方法可以有效识别早期故障特征，具有较高的故障识别率。第三，时间序列状态预测技术研究。故障预测是PHM的关键，针对渐变性故障类型，提出基于自回归滑动平均模型(Auto-Regressive and Moving Average Model，，ARMA)和人工神经网络（Artificial Neural Network，ANN）的状态预测方法。利用ARMA模型在捕捉时间序列线性部分的优势和ANN处理非线性时间序列的良好性能，建立混合预测模型。采用ARMA模型建模时间序列的线性部分，通过序列余项建立神经网络模型，实现非线性部分建模。基于该混合模型，以某型号卫星遥测电压数据为例，实现对遥测参数的静态模型预测和动态模型预测，实验结果表明混合动态模型具有较高的预测精度。基于混合模型，以遥测压力数据为例实现了一种状态监测方法，实验表明该方法可以有效降低故障虚警率。第四，故障知识建模及服务技术研究。故障知识管理是PHM的基础，针对目前故障诊断知识资源重用率低、知识断层严重等问题，提出基于领域本体的故障知识建模和服务技术。首先，提出静态和动态两类故障知识分类方法，构建了推理知识本体和多领域知识本体；然后，基于推理知识本体实现了故障知识推理，基于多领域本体，实现了对分布式知识资源的索引、语义标注、语义检索及知识管理，有效获取产品设计、试验等阶段的诊断静态知识；通过对相对固化的诊断、预测模型类知识资源进行服务化封装和语义标注，实现动态知识的检索和调用服务，有效提高诊断知识资源的使用效率和重用率。第五，PHM服务系统开发。基于以上研究内容，以航天器为例，针对其在轨运行特点，研发航天器地面PHM系统。集故障预测、故障诊断、故障知识管理和服务于一体，为提高航天器在轨运维保障能力和地面测试试验能力提供有效的技术支撑。本文系统地分析和研究了复杂装备PHM系统的服务模式及关键技术，并结合航天器特点，开发了相应的PHM原型系统，验证了本文所提方法的可行性和有效性。通过本文研究，对提高复杂装备诊断维护能力、降低运维成本具有重要意义。
[Abstract]:In recent years, the safety and reliability of complex equipment systems such as aviation, space and ship have become the focus of attention. As a result of its complex structure, a large loss can be caused once a failure occurs. Therefore, it is urgent to improve the reliability, repair and safety of the complex equipment system. However, the current research work in the field of fault diagnosis is mainly focused on the state evaluation and fault diagnosis of the research system, and concerned is the running state of the system "Current", and the research on system fault prediction and health management is less. This traditional "post-post service" and "plan and repair" approach is seriously inadequate in dealing with new situations of transient change. The "visual maintenance" and the "Predict maintenance" eliminate the fault in the bud state and become the development direction of the system maintenance and guarantee in the future. Therefore, in recent years, the research on the fault prediction and health management (PHM) has become the research focus of the domestic and foreign scholars, and the PHM technology represents the fault diagnosis technology from the traditional sensor-based diagnosis to the prediction of the intelligent system. But, at present, the research in this area has just started, and the research results are more In this paper, the key theory and technology involved in the PHM technology of complex equipment are studied, and the main content package includes: first, complex equipment PHM service mode Based on the analysis of the characteristics of complex equipment and the shortcomings of the traditional fault diagnosis model, a new service-oriented PHM system is proposed. This paper studies the system service model and characteristics, analyzes the key technology and main function modules required in the system construction process, and provides a kind of public PHM system service platform research in the field of complex equipment. New ideas. Second, data reduction and failure diagnosis Fault diagnosis is the core content of PHM, which is based on nonlinear manifold learning and Hidden Markov Model (HMM). The hybrid HMM model divides the system state into the normal state, the intermittent fault, the intermediate state and the fault state, so as to fully reflect the system state. In this paper, the original high-dimensional fault data is mapped to a low-dimensional space by a local holding projection algorithm (LPP), and the inner manifold characteristic of the extracted data is taken as a feature vector, and the mixed HMM is used as a classifier to realize the pair of states. The results show that the LPP-HMM method can effectively identify the early fault features and has higher accuracy. Fault recognition rate. Third, time series The prediction of fault is the key to the PHM, and the auto-regressive and moving average model (ARMA) and the artificial neural network (ANN) are put forward. The method of state prediction is based on the advantage of the linear part of the capture time series and the good performance of the ANN processing non-linear time series by using the ARMA model. The linear part of the time series is established by the ARMA model, and the neural network model is established by the remainder of the sequence. Based on the hybrid model, a model of satellite telemetry voltage data is taken as an example to realize the static model prediction and the dynamic model prediction of the telemetry parameters. The experimental results show that the hybrid dynamic model has the characteristics of Based on the mixed model, a state monitoring method is realized by using the telemetry pressure data as an example, and the experiment shows that the method can be used effectively. Reduce the false alarm rate of the fault. Fourth, the fault knowledge Modeling and service technology research. The fault knowledge management is the foundation of the PHM, aiming at the problems of low reuse rate, serious knowledge fault and the like of the current fault diagnosis knowledge resource, and the fault knowledge management is based on the fault of the domain ontology in that first place, two types of fault knowledge classification method of static and dynamic are put forward, and the reasoning knowledge body and the multi-domain knowledge body are constructed; then, based on the reasoning knowledge body, the fault knowledge reasoning is realized, The index, the semantic annotation, the semantic retrieval and the knowledge management of the source can effectively acquire the diagnosis static knowledge in the stages of product design, test and the like; and through the service encapsulation and the semantic annotation of the relative curing diagnosis and the prediction model knowledge resources, the dynamic knowledge is realized. Knowledge search and call service to effectively improve the diagnosis knowledge resource Use efficiency and reuse rate. 5. The development of PHM service system. Based on the above research contents, taking the spacecraft as an example, for the characteristics of its on-orbit operation, the research The ground PHM system of the spacecraft is developed. The fault prediction, the fault diagnosis, the fault knowledge management and the service are integrated to improve the capability of the spacecraft in the operation and maintenance of the rail and the ground test. In this paper, the service mode and key technology of the PHM system of complex equipment are analyzed and studied systematically, and the corresponding PHM prototype system is developed and verified by combining the characteristics of the spacecraft. The feasibility and effectiveness of the method proposed in the paper. Through the research of this paper, it is necessary to improve the diagnosis and maintenance of complex equipment
【学位授予单位】：北京理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TH165.3

【相似文献】