【作者】 蒲星星；

【导师】 蒋洪德；

【作者基本信息】 清华大学 ， 动力工程及工程热物理， 2013， 博士

【摘要】 燃气轮机气路故障不仅造成燃机性能下降，影响燃机的经济性，气路故障如果不能被及时发现和维修，还会影响燃气轮机机组运行的安全性和可靠性。气路故障诊断对燃机部件气路通流部分的运行状态进行实时的监测和评价，针对燃气轮机的运行状态合理安排燃机的运行和维修计划，减少不必要维修，最大限度的提高燃气轮机系统的安全性、可靠性和经济性，因此燃气轮机气路故障诊断研究具有重要的理论意义和工程应用价值。本文重点研究了基于模型的重型燃气轮机气路故障诊断方法，基于此建立了基于模型的重型燃气轮机气路故障诊断系统。首先，建立面向对象的重型燃气轮机自适应气路故障诊断模型。将模型仿真结果与燃气轮机运行实测数据对比验证所建立模型的准确性。建立重型燃气轮机状态空间模型、线性和非线性自适应气路故障诊断模型以及基于扩展卡尔曼滤波器的自适应气路故障诊断模型，用于重型燃气轮机气路故障仿真和诊断。其次，研究基于模型的重型燃气轮机气路故障诊断精度评价和测量参数优化选择方法。提出一种新的状态能观度分析方法，高能观度状态和高总体能观度测量组具有高故障诊断精度。利用能观度方法分析测量噪声、冗余测量、燃气轮机非线性对燃气轮机各状态能观度和故障诊断精度的影响；分别对基于动态模型和稳态模型的重型燃气轮机气路故障诊断方法进行测量参数优化选择。再次，研究基于自适应扩展卡尔曼滤波器的重型燃气轮机气路故障过程跟踪。针对扩展卡尔曼滤波器算法的不足，提出基于强跟踪滤波器的燃气轮机自适应气路故障诊断方法，实现对燃气轮机渐变故障和快变故障的准确跟踪。基于卡尔曼滤波器原理对强跟踪滤波器算法进行改进，提高算法对故障幅值、滤波器初值以及测量噪声的鲁棒性。最后，研究基于模型的燃气轮机欠定气路故障诊断问题，提出基于稀疏贝叶斯学习算法的欠定气路故障诊断方法，用于燃气轮机快变气路故障和传感器故障同时诊断。通过对大量故障诊断实例分析，验证方法的有效性、准确性和较已有方法的优越性。更多还原

【Abstract】 The unscheduled maintenance or outages of gas turbines due to degradation cannot only induce great additional costs and lost revenues but also influence the safety andreliability of gas turbines. Gas path diagnostics and prognosis can be used to assessengine conditions in real-time and predict the failure time in the future, so the predictivemaintenance actions can be performed. The benefits of predictive maintenance actionsare improved usability, safety and reliability as well as reduced life cycle costs.Therefore, the study of gas path diagnostics has great theoretical significance andpractical value. This dissertation focuses on the research on the model-based gas pathdiagnostic method, and develops model-based gas path diagnostic system of heavy-dutygas turbine.Firstly, an adaptive generic object-oriented model for gas path diagnostics ofheavy-duty gas turbine is established. The model simulation results match with the fielddata, which confirms the accuracy of the model. The state space model, linear andnonlinear adaptive gas path diagnostic model, and Extended Kalman Filter(EKF) basedadaptive gas path diagnostic model of heavy-duty gas turbine are developed to analyzeand diagnose the gas path fault.Secondly, the assessment of gas path diagnostic accuracy and optimalmeasurement selections method for model-based gas path diagnostics is studied. Anovel degree of observability analysis method for measurement selections of gas pathdiagnostics is developed. The states with high degree of observability and themeasurement sets with high overall degree of observability result in high estimationaccuracy in gas path diagnostics. Using the proposed method, the influence of themeasurement noise, the overdetermined measurement, the gas turbine nonlinearity andthe advanced measurement such as turbine inlet temperature on degree of observabilityand gas path diagnostic accuracy are analyzed. The optimal measurement selections areconducted for dynamic and stable model-based gas path diagnostics.Thirdly, the tracking of gas path fault based on adaptive EKF is studied. Accordingto the drawbacks of the EKF, adaptive gas path diagnostics using the strong trackingfilter(STF) is proposed. The proposed method can track both the abrupt fault andgradual fault accurately, which overcomes the drawbacks of the EKF. Based on the analysis of the principle of the EKF, the algorithm of the STF is improved to increasethe robustness against the gas path fault magnitude, initialization value of filter andmeasurement noise.Finally, the underdetermined gas path diagnostic problem of model-based gas pathdiagnostic method is studied. A new gas path diagnostic method based on SparseBayesian Learning favoring sparse solutions for abrupt fault events is proposed, whichcan be used to diagnose the component fault and sensor bias fault together for abruptfault events. The analysis of the a variety gas path diagnostics results demonstrate thecapability, accuracy and superior over other methods.更多还原