【作者】 董晓峰；

【导师】 顾煜炯； 杨昆；

【作者基本信息】 华北电力大学 ， 热能工程， 2012， 博士

【摘要】 故障诊断在汽轮机组状态维修中具有重要的地位,故障诊断系统应该作为状态维修决策支持系统的一部分,为维修决策提供技术支持,如何将RCM()分析结果应用到故障诊断过程中,使故障诊断为状态维修提供更有价值的信息；如何将不断发展的智能故障诊断技术应用于汽轮机组故障诊断,以提高汽轮机组故障诊断的准确性和可靠性,都是需要进一步研究的问题。对汽轮机组振动故障和通流部分故障进行RCM分析,充分了解这些故障产生的原因,造成的影响,需要采取的故障处理措施,故障发生时都有哪些故障特征以及如何进行监测以获取这些故障特征。应用RCM分析结果,建立故障诊断模型和故障诊断流程,应用故障诊断流程调用故障诊断模型对故障进行诊断,得到故障的故障模式、故障原因、故障位置、故障影响和故障处理措施等诊断结论,形成故障诊断报告,为维修决策提供技术支持。应用故障特征对故障模式进行聚类分析,形成故障模式类,从而可以在故障模式类层次区分开属于不同性质的故障模式,解决类间的识别问题,进而缩小故障诊断的识别范围。在故障模式类中使用粗糙集方法从故障特征中提取对故障诊断有贡献的故障特征,排除冗余的故障特征,减少冗余故障特征对故障识别的干扰,优化故障诊断规则。在诊断推理过程中,需要给诊断规则中的前提条件赋予相应的权重,由专家凭经验给出权重的方法因其具有一定的主观性而影响了权重对客观实际的反映,从而增加了诊断结果的不确定性,应用粗糙集理论中知识依赖度得到的权重,克服了主观分配权重存在的不足,使权重分配结果更符合客观实际。应用统计分析方法来求取故障监测参数的正常工作范围,结合运行规程的规定确定故障监测参数各段工作范围的隶属度,选择合适的函数作为故障征兆的隶属度函数。计算多个工况下汽轮机通流部分故障监测参数的正常工作范围,运用曲线拟合得到工况与故障监测参数正常工作范围的映射,解决了变工况下故障征兆隶属度函数的自动获取问题。以现场应用为目的,利用最新的软、硬件技术,参与设计并开发适用于状态维修要求的智能化汽轮机组故障诊断系统。更多还原

【Abstract】 Fault diagnosis is important for turbogenerator condition maintenance because fault diagnosis provides technical support for maintenance decision. How to put the result of reliability centered maintenance analysis into the process of fault diagnosis to make fault diagnosis provide more important information for condition maintenance and how to use advanced intelligent fault diagnosis technology for turbogenerator fault diagnosis to improve the accuracy of fault diagnosis result are study content in this dissertation.Reliability centered maintenance analysis method is used to analyze steam turbine vibration fault and flow path components fault in order to get some information of fault reason, fault influence, fault treatment measures and fault symptoms about the fault. The information is used to build fault diagnosis model and fault diagnosis flow. The fault diagnosis model is used to make fault diagnosis follow fault diagnosis flow. At last, fault diagnosis report is gotten to provide technical support for maintenance decision.With the purpose of reducing fault identification range, extracted principal component feature of fault is used to class fault modes into several fault classifications in order to identify the fault classification to which the fault belongs in the first step for fault diagnosis. The fault diagnosis decision table is built by using rough set theory, which is used to extract fault symptoms that are useful for identifying the fault from several fault modes included in a fault classification. So the fault diagnosis rules are optimized by this method, the other reductant rules are dismissed to descrease the influence to fault identification.The weight distribution method by expert experience is often used in the process of fault diagnosis inference to distribute the weight of the precondition of fault diagnosis rules. This method has the disadvantage of making the weight with the influence of human subjectivity. A method of weight distribution using knowledge dependency which is used to overcomes the disadvantage of weight distribution using expert experience to decrease the uncertainty in the inference process of fault diagnosis.The normal operation range of thermal parameters is obtained by using a statistic analysis method, which combined with operation rules to obtain all kinds of operation range membership degree of thermal parameters. After then, membership function of fault symptoms are got by selecting suitable function. A mapping between normal operation range of a fault systom and working condition under multi-load is built, which to solve the problem of the fault symptom membership function’s automatic acquisition.With the purpose of field application, the intelligent fault diagnosis system for turbogenerator was developed to meet the requirement of condition maintenance by using latest technology of hardware and software.更多还原