【作者】 黄强；

【导师】 吴建军；

【作者基本信息】 国防科学技术大学 ， 航空宇航科学与技术， 2012， 博士

【摘要】 论文以我国高压补燃液氧煤油发动机为研究对象，针对其健康监控所涉及的故障仿真、传感器配置优化、传感器故障检测与数据恢复、实时故障检测与诊断、系统集成设计与实现验证，以及发动机飞行前、飞行中和飞行后健康监测等关键技术开展了全面深入的分析、设计、实现等研究工作。研究结果不仅为研制工程实用的一次性使用发动机健康监控系统奠定了坚实的理论和技术应用基础，而且对提高我国未来可重复使用液体火箭发动机的可靠性、安全性具有重要的参考价值。针对发动机工作条件的极致性、故障发生发展的快速性、影响后果的严重性、故障模式的可复制性差，从而导致发动机故障模式特征、诊断知识和样本数据难以获取等问题，基于结构层次化分解的思想，建立了模块化的发动机故障仿真模型，开发实现了发动机可视化故障仿真软件系统，对发动机氧化剂泵汽蚀、燃烧室喉部烧蚀等主要故障进行了故障仿真及效应分析。仿真计算结果与发动机实际试车数据吻合较好，可以为故障检测与诊断方法提供重要的发动机故障样本数据。针对液体火箭发动机故障检测与诊断在缺乏先验知识、缺少充分样本数据条件下的不确定性信息决策问题，基于有机结合随机性和模糊性的云理论，深入开展了检测参数选择与传感器优化配置、传感器故障检测和数据恢复、发动机工作过程实时故障检测、故障诊断的不确定性推理方法研究。在保证对发动机故障模式分类能力不变的情况下，研究发展了发动机故障检测参数选取方法，同时将故障检测与诊断性能指标作为约束条件，建立了发动机基于云理论的传感器配置优化数学模型，并结合粒子群算法研究了发动机传感器配置优化问题。发展了基于云－神经网络的发动机传感器故障检测与数据恢复方法，结合某型液氧煤油发动机试车数据进行了实例分析与验证。结果表明，该方法可行有效，可以为发动机的故障检测与诊断提高可靠的信号数据源。针对液体火箭发动机实时故障检测在准确性、及时性和实时性的要求，并结合云理论和神经网络强大的数据处理能力，提出了液体火箭发动机故障检测的一种云－神经网络原理结构，发展了云－神经网络的前向传播计算和反向传播学习算法，实现了液体火箭发动机工作全过程基于云－神经网络的实时故障检测方法。针对某型液氧煤油发动机起动过程、额定工况到高工况过程、高工况到高工况高混合比过程、高工况高混合比到高工况过程等瞬变过程，以及额定工况、高工况和高工况高混合比等稳态过程的实际试车数据，对该方法进行了验证。验证结果表明，该方法可以对发动机工作状况进行及时的判断，没有误报警和漏报警，且相比RS、IATA、ACA、RBF等故障检测算法能更早地检测出故障。综合云理论和Petri网在描述分析发动机动态行为与状态变迁过程方面的能力和特点，建立了液体火箭发动机故障诊断的云－Petri网模型，发展了基于规则的云－Petri网建模方法，实现了液体火箭发动机基于云－Petri网的故障诊断方法，并利用某型液氧煤油发动机试车数据对诊断方法进行了实例分析和验证。结果表明，该方法可以对发动机氧涡轮泵前管路堵塞、氧涡轮泵汽蚀等典型故障进行隔离与诊断。针对当前液体火箭发动机健康监控系统设计开发过程中普遍存在的系统结构和功能模块紧耦合、重用性和互操作性差、难于快速响应系统后期的需求变化和维护等诸多缺陷，在深入分析液体火箭发动机健康监控系统功能与需求的基础上，将发动机健康监控所面临的共性问题进行提炼、抽象，分析设计了基于数据-模型-控制-视图的分层、开放和可复用的发动机健康监控系统框架。同时，针对我国某型液氧煤油火箭发动机，结合相关研究成果，设计实现了其工作过程的实时故障检测系统，并开展了该系统基于发动机试车数据和地面功能试车的考核与验证。结果表明，所设计的系统完全满足工程实用的需要，具备实时在线运行能力，不仅能实现参数的准确采集，而且算法没有出现误报警和漏报警。针对发动机重复使用过程中对健康监控技术提出的更高要求，开展了发动机飞行前综合性能测试、飞行过程实时状态记录和飞行后结构检测等健康监测技术的原理性分析与初步概要设计。分析设计了发动机飞行前地面综合性能自动测试系统和发动机智能机内测试系统原理结构；结合光纤光栅等先进测量传感技术，分析设计了由辅助功能、系统控制、数据采集与存储、回收分析等功能模块组成的发动机飞行参数记录仪原理结构；分析设计了发动机飞行后关键构件基于内窥图像获取和结构损伤检测与识别的内窥检测系统原理结构。结果对研制和开发我国未来可重复使用液体火箭发动机飞行前、飞行中和飞行后等阶段的健康监测技术与工程实用系统，具有重要的参考价值。更多还原

【Abstract】 Some key technologies involved in the health monitoring for a certain high pressurestaged combustion LOX/Kerosene rocket engine have been comprehensively studied,analyzed and designed in the thesis. These include the configuration, optimization, faultdetection and data accommodation of sensors, the fault simulation, real-time fault detectionand diagnosis methods for the engine, the integrated design, implementation and validationof the system, and the health inspection technique in the preflight, inflight and postflightphases of reusable engines. Results of the research can not only provide solid theoreticaland application foundation in the development of health monitoring system for expendablerocket engines, but also they can provide important reference in the improvement ofreliability and safety for future reusable liquid-propellant rocket engines.Firstly, due to the extreme operating condition, rapid occurrence and severe effect offaults, and the bad repetition of fault modes, it is difficult to obtain the feature of faultmodes, the diagnostic knowledge and the sample data of engines. So modularizedsimulation models of faults are founded and a visual fault simulation software system isdeveloped based on the hierarchical decomposition of the engine’s structure. Then faultsimulation and effect analysis are carried out for the principal faults such as the cavitationin the oxidant pump, the ablation of the combustion chamber throat, and so on.Comparison results show that the simulation results are in good agreement with the realtest data of the engine. It can provide important fault sample data in the research of faultdetection and diagnosis methods for the engine.Secondly, due to the intrinsical uncertainty existed in the fault detection and diagnosis,especially in the absence of prerequisite knowledge and sufficient sample data, uncertainreasoning and decision methods are proposed and studied. The methods are all based onthe cloud theory, which is effective in the integrated description and manipulation of therandomness and fuzziness. They include the following three aspects such as the parameterselection, configuration, fault detection and data accommodation for sensors, the real-timefault detection method and the uncertain diagnosis reasoning method for the engine.In the first aspect, a parameter selection method for fault detection with unchangeableability in the classification of the engine’s faults is developed. Then a mathematical modelof configuration and optimization for sensors is built based on the cloud theory to satisfythe performance index constraint of fault detection and diagnosis. A computation method isalso proposed based on the particle swarm optimization. In addition, a fault detection anddata accommodation method for sensors is developed based on the cloud neural networkand verified with test data of the LOX/Kerosene rocket engine. Results show that themethod is effective. By this way, reliable data for the fault detection and diagnosis of theengine can be obtained. In the next aspect, a real-time fault detection method for LRE is proposed based onthe cloud neural network to meet the accurate, in-time and real-time requirements. For themethod, a framework and structure of the cloud neural network are designed, and theforward propagation based computation and the backward propagation based learningalgorithms are developed. Then the method is verified and validated with test data of theLOX/Kerosene rocket engine in both of its steady processes and transient processes. Thesetransient processes include a starting process, a process from the rated condition to the highoperating condition, a process from one high operating condition to another high operatingcondition with high mixing ratio, a process from one high operating condition with highmixing ratio to one high operating condition. These steady processes include a process ofrated condition, a process of high operating condition, and a process of high operatingcondition with high mixing ratio. Results show that the developed method can not onlyproduce right and timely estimations for operating condition of the engine with no falsealarm and missing alarm, and also it can detect the fault earlier than other fault detectionalgorithms such as the RS, IATA, ACA and RBF.In the last aspect, the cloud Petri net model of fault diagnosis and its modeling methodbased on the productive rules are proposed. The model is used to describe and analyze theengine’s behavior and the change between states effectively. Then an uncertain reasoningmethod for the fault diagnosis and discrimination of its reasoning results is proposed anddeveloped for the engine based on the constructed cloud Petri net model. The method isverified and validated with test data of the LOX/Kerosene rocket engine. Results showthat the method is able to isolate and diagnose the typical faults of the engine such as theblock of pipeline before oxidizer turbopump, the cavitation of oxidizer turbinepump, andso on.Thirdly, the function and requirements of HMS for LRE are analyzed thoroughly, andthe common problems are also abstracted due to the many deficiencies existed in thecurrent HMS such as tight coupling of system structure and function modules, poorreusability and interoperability, the slow response for the demand change and maintenance.Then a layered, open and reusable framework of HMS is analyzed and designed based onthe Data-Model-View-Control hierarchy. Moreover, based on the combination of theprevious research results, a real-time fault detection system for the LOX/Kerosene rocketengine is designed and realized. The system is verified and validated with test data andground tests of the engine. Results show that the system can meet all the requirements inthe engineering application and is able to work in a high real-time and on-line way. It cannot only acquire the parameters accurately, and also produce no false alarm and missingalarm. So it is another important breakthrough in the research of health monitoring forLRE in China.Fourthly, in order to meet the higher requirement of health monitoring for reusable engines, key health inspection technologies are analyzed on principle and designedsummarily such as the preflight integrated performance test, real-time flight data recordingand postflight structure detection. The principle structure of automatic test systems for theon-ground integrated performance test and intelligent Built-in Test in preflight phases ofengines are analyzed and designed. Then the principle structure of real-time flight datarecorder is analyzed and designed which is combined with fiber gratings and otheradvanced sensor technology. It is composed of auxiliary function module, system controlmodule, data acquisition and storage module and recall-bach module. Moreover, theprinciple structure of borescopic inspection detection system is also analyzed and designedbased on image acquisition and detection and identification of structural damage. Theseresults will provide significantly important reference in the research of health inspectiontechnology and development of engineering application systems for the preflight, inflightand postflight phases of future reusable LRE.更多还原