【作者】 张彼德；

【导师】 孙才新；

【作者基本信息】 重庆大学 ， 电气工程， 2002， 博士

【摘要】 随着汽轮发电机组容量不断增大，结构越来越复杂，各个零部件之间的联系更加紧密，虽然引起机组振动的原因也更趋多样化，复杂化，但振动状态仍是衡量机组运行状况好坏的一个重要标志。同时，由于汽轮发电机组是一个多故障系统，经常出现多种故障同时发生的情况，而目前对于多故障的诊断问题还未能很好的解决。因此，研究汽轮发电机组振动故障诊断技术既对机组的安全稳定运行有着十分重要的意义和巨大的经济效益，也是电气工程等多学科交叉的前沿研究课题。本文在总结和借鉴前人关于汽轮发电机组状态监测和故障诊断的基础上，重点研究基于模糊集理论、概率因果模型、人工神经网络技术以及遗传算法的多故障诊断理论和方法，并用已有的汽轮发电机组振动故障事例进行了验证，得出了一些具有实用价值的结论，进一步丰富和推进了大型汽轮发电机组振动多故障诊断的理论，并提出了一些可靠的、实用的新方法。 众所周知，汽轮发电机组振动故障诊断技术是一项非常复杂的系统工程，涉及的相关学科领域较多，与现场实践经验联系十分密切，它一直是工程界十分关注和重视的课题。本文的主要内容及创造性成果有： (1)通过对模糊综合评判方法在汽轮发电机组振动故障诊断中应用的综合分析研究，提出了一种适用于汽轮发电机组多故障诊断的分层模糊诊断模型。 (2)运用模糊聚类分析理论，将标准故障样本和侍检数据一起作为分类样本，利用基于模糊等价矩阵的传递闭包法，提出了一种用于汽轮发电机组多故障诊断的新方法。 (3)通过概率因果模型与节约覆盖集理论关系的分析，建立了一种用于汽轮发电机组振动多故障诊断的概率因果模型和基于概率因果的联接模型。 (4)利用概率因果模型中征兆与故障之间的关系，对BP网络进行了改进，增加了输入与输出相关节点的直接联接。 (5)建立了汽轮发电机组多故障诊断的两级神经网络模型，它不仅对多故障诊断识别精度高和模块化程度高，且具有发现新故障的能力。 (6)通过对自组织特征映射神经网络聚类功能的研究，将它与模糊聚类相结合实现了对汽轮发电机组的多故障诊断。 (7)通过模糊c-划分极易陷入局部极小值和遗传算法有全局寻优特点的分析，提出了将两者相结合的聚类分析新方法，不仅可克服模糊c-划分陷入局部极小点的缺陷，而且能对汽轮发电机组的多故障进行有效的诊断。 (8)针对概率因果模型对于节约原则的选取没有一个定量的标准和故障数目很 奎庆大学博十学位论文 多时问题解的“组合爆炸”问题，提出了一种将遗传算法和概率因果模型 相结合的故障诊断方法。 （9）提出了以遗传算法对神经网络的权值和阈值进行优化并用神经网络细调的 多故障诊断方法。 本文提出的上述模型及方法，经实例分析验证，可对汽轮发电机组振动多故障进行有效诊断，在一定程度上丰富了机组多故障诊断技术的理论及方法，有良好的应用前景。更多还原

【Abstract】 With the increasing capacity of steam turbine generator-set and the complexity of its structure,the relations among components are getting closer than ever. Although the causes resulting in a machine set’s vibration become more various and complex,the vibrant state is still a very important indication to judge the working conditions of machine set. At the same time,steam turbine generator-set is a system,which often has multiple faults synchronously,and the problem for the multiple faults has not been solved appropriately. Therefore,the research on vibration fault diagnosis not only has great importance and benefit for a machine set to work safely and stably,but also is a frontier topic for electrical engineering and other cross fields. Based on predecessors’ studies on steam turbine generator-set’s state monitoring and fault diagnose,this paper focuses on the models and methods used to judge the multiple faults. These models and models,which are based on fuzzy sets theory,probability causal model,artificial neural network technology and genetic algorithms,are tested by some existing examples and some useful conclusions are drawn. All above will put the theory of multiple fault diagnosis of a large turbine generator set a further step.It’s well known that the vibration fault diagnosis technology on steam turbine generator-set is a very complicated system project,which involves many fields and has a close relationship with practice. So it always draws much attentions from engineering field.The main contents and the creationary outcomes of this paper are described as follows:(1) The limitation of the fuzzy general judge is found by analyzing its application in steam turbine generator-set’s fault diagnosis and a multi-layer fuzzy diagnosis model which can overcome these limitations is put forward.(2) A new method,which can be used to diagnose steam turbine generator-set’s multiple faults,is brought out. This method,which is based on fuzzy clustering analyze theory,puts standard fault samples and the checked samples together as classified samples and draw a conclusion by using transfer closure based on fuzzy equivalence matrix.(3) The probability causal model and the connective model based on the probability causal for the fault diagnosis of steam turbine generator-set are put forward by the analysis of the relationships between the probability causal model and theparsimonious cover set theory.(4) Applying the relationships between faults and fealtures in probability causal model,the BP neural network is improved by adding the direct connections between the input nodes and the output nodes.(5) A two-layer neural network model for steam turbine generator set’s multiple fault diagnosis is put forward. It can identify the multiple faults and has high modularization. Also it can find the new fault.(6) By the research of clustering capability of SOM,associated with fuzzy clustering analysis,we can realize the diagnosis of multiple faults of turbine generator-set.(7) By the analysis of the fuzzy-c partition and the genetic algorithm,a new method combing with them is put forward. This method can prevent the fuzzy-c partition from the local least point and can diagnose the steam turbine generator-set’s multiple faults more effectively.(8) Considering that probability causal model has not a qualitative standard to select parsimonious rules and there could be a problem of "combination explosion" when there are so many faults in the model,a method which combines probability causal model and genetic algorithm is put forward.(9) A method,which combines the neural network and the genetic algorithms,is advanced. It uses the genetic algorithms to improve the weight and threshold of the neural network first. Then using the neural network,the diagnosis result can be gotten.The methods and models that this paper put forwards can diagnose the vibration multiple faults of steam turbine generator-set effectively,and this could be proved by examples. It enriches the theory and更多还原