【作者】 张可；

【导师】 柴毅；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 随着现代工业及科学技术的迅速发展,机械设备和工业系统日趋大型化、网络化和自动化,功能不断增多,结构愈加复杂,设备系统故障的相互耦合和影响以及多故障特性,对传统的故障诊断技术如何适应复杂设备系统实际安全运行的需要提出了挑战。由于多故障自身的复杂性和不确定性,一种故障可能由多个原因引起,而一个原因又可能引发多种故障,从而成为一种多对多的复杂映射关系。不同的故障和征兆构成了不同的集合或关联域,使得多故障的诊断成为了一个复杂的问题。在一般以单故障诊断方法应用于多故障的系统中,诊断结果与实际故障存在着较大的差异。因此,需要深入分析征兆和多故障的描述和映射关系,研究多故障及其征兆的智能故障诊断方法。论文分析了当前多种故障诊断方法,以及在单一故障和多故障诊断中的应用。指出了当前多故障诊断存在的不足、已有方法在多故障诊断中的局限性、以及研究多故障诊断的意义。讨论了故障征兆的描述和多故障的模型描述,根据实际故障体系中故障和征兆多对多的复杂对应关系,研究适用于多故障诊断的故障征兆矢量及其描述,构建了“征兆—故障”映射关系。同时通过不同故障体系中征兆对于多故障的表现,将多故障可由组成其的单一故障的征兆表示的,定义为可分离型多故障;而组成多故障的征兆相互耦合,难以按一般数学方法解耦的成单一故障的,定义为征兆—故障耦合型的多故障,对于这种多故障,本文基于人工神经网络进行了诊断的研究。论文按照多故障的征兆特征表现,建立了故障征兆矢量的空间描述,以空间欧氏距离、相似程度、匹配度等指标来作为诊断的依据,同时讨论了归一化和模糊控制在故障征兆矢量取值中的应用。对于“可分离型”多故障,分别讨论了基于逻辑运算、相关系数、征兆的故障概率的多故障诊断方法,该三种方法均是通过数学计算,对征兆进行分析,直观的将待诊断故障与已知故障之间的关系和差异用数值表示出来,以此作为多故障诊断的重要依据和参考.针对多故障复合发生诊断的问题,提出了一种基于征兆邻搜索优化聚类的自组织映射多故障诊断算法,建立了具有三级分析结构的SOM网络的多故障诊断模型;给出了一种基于半径搜索的优化算法,以空间最小距离对故障征兆索引集的邻搜索进行了优化和改进,减小了聚类偏差,使得优化后聚类结果的准确率得到较大的提升,突出了SOM网络自组织映射的能力。仿真及应用结果表明优化后提高了多故障诊断的准确率。针对离散型征兆的多故障,提出了基于径向基函数与自联想记忆神经网络结合的诊断算法,利用Hopfield网络将征兆与神经元映射记忆样本故障,使网络状态与故障空间对应;根据Hopfield网络的联想记忆能力,建立故障征兆组合和故障间的映射关系并进行辨识,通过网络的反馈将待诊断故障进行联想匹配,利用RBF网络将匹配后的征兆与故障映射,结果表明有效的实现多故障诊断。提出了一种改进Hopfield网络的异联想记忆多故障诊断算法,将故障及其征兆分别对应到网络输出和输入,构造联想记忆权值矩阵,并建立网络最小偏差的优化求解方法,使网络输出趋向能量最小的顶点,实现单一故障或多故障的诊断。算法对于输入错误的状态数据有一定的容错性,同时可以方便的对方法进行扩展运算速度快,仿真及应用结果表明该算法对于多故障诊断具有较高的准确性。在实际获取到的数据基础上,分析了故障诊断和病害诊断的区别和联系。验证了本文所提出的多种方法的通用性。仿真结果显示本文提出的几种多故障诊断方法能够模拟人的思维模式,较为有效的识别汽轮发电机组故障体系和番茄病害体系中的单一故障(病害)和多故障(病害),提供诊断策略支持,符合故障诊断的需求。最后讨论了本文提出的各个方法的优缺点和适用范围,对全文的研究工作进行了总结,并进一步指出了工作的研究方向。更多还原

【Abstract】 With the rapid development of modern industry and sci-technology, the industry equipments increasingly become large-scaled, continued, networked, high-speeded and automated, which accordingly leads to more and more functions, and more and more complicated structures. Meantime, multi-fault occur simultaneously in mechanical equipment has become frequently, which requires the new development of intelligence fault diagnosis.Due to single fault may be caused by many reasons, and one reason may cause many faults. Thus, there is a need to form a multiple-to-multiple map. As different faults and symptoms form different aggregations or regions, the multi-fault diagnosis has become a very complicated task. In traditional intelligence fault diagnosis process, the characters of multi-fault are not deeply analyzed and processed. The methods and systems designed to diagnose multi-fault are incomplete, the systematic research is not yet developed widely. Therefore, there are differences between diagnosis results and actual faults in ordinary fault diagnosis system. More researches for designing a multi-fault intelligence diagnosis method and symptoms based multi-fault description are needed. Especially, researches on the description and diagnosis of symptoms are needed as well.Many fault diagnosis methods and applications are analysed in this paper. Aiming at the fault that its symptom is numerical or can be converted to numeral, the definition of multi-fault diagnosis is proposed. At the same time, the deficiencies and limitations of existing multi-fault diagnosis methods are analysed. The relationship between multi-fault diagnosis and mono-fault is discussed, and the research direction and route are proposed. The main research results are:Both description of fault symptom and multi-fault model are discussed in this paper. Focused on mapping realtionship“multi to multi”between fault and symptom in actual fault system, conducted researches on fault symptom vector and the description for multi-fault diagnosis, the mapping relationship“symptom to fault”is proposed. Meanwhile, according to symptom indications in different kinds of multi-fault, defined multi-fault composed of several symptoms of single faults as separable multi-fault. On the other hand, for multi-fault that composed of the coupled symptoms, hard to be decoupled by normal mathematic methods are defined“symptom couple”multi-fault, artificial neural network (ANN) based researches is conducted as well.According to description of multi-fault symptom, symptom is mapped in fault-space, the space-indication of symptom vector is established. Euclidian distance, similarity and marching score are taken as diagnosis coordinates. Application of normalization and fuzzy control for dereferencing fault symptom vector is discussed. For“separable”multi-fault, diagnosis methods based on logic mathematical, correlation coefficient, probability calculation are proposed. Symptom analysises of all of the three methods are using methmatical calculation, the relationship and difference between actual fault and known faults are shown in numerical vaule intuitively, which can be used as important factor and refenrence.A multi-fault diagnosis method based on self-organizing map (SOM) and optimization of the adjacent-searching is developed. A SOM neural network that used to multi-fault diagnosis is created. The inaccurate clustering of traditional SOM algorithm is analyzed. According to the analysis, Euclidean distance is taken as the main discrimination, and the adjacent-searching algorithm is optimized. Using the optimized algorithm, the cluster results of input samples are obtained, symptomss of faults are mapped, and a multi-disease diagnosis model is developed. The proposed SOM-based model has three layers. The symptom array of faults can be accurately sorted and clustered using the optimized model.A multi-fault diagnosis method based on combining radial base function (RBF) and self-associational memory is developed. Aiming at a problem that the medium Hopfield neural network only has self-associational memory property, according on proposed multi-fault diagnosis method bases on self-associational memory and fuzzy control, RBF neural network is used to improving the method. The associational memory capability of Hopfield neural network, the adaptive capacity and self-learning capacity of RBF nerual network are used to memorizing and classification. Through the implementation of the process, all of the symptom and fault are classified and mapped, and multi-fault diagnosis is achieved.A multi-fault diagnosis method based hetero-associative memory is developed. Taking bidirectional associative memory neural network as reference, a Hopfield network that has hetero-associative memory property is proposed. Four weight matrix building method are discussed.Based on the actual acquired data, the relationship and deference between fault diagnosis and disease diagnosis are annlysed. The commonality of the multi-fault diagnosis methods proposed in this paper is proved. The turbine generator fault system and tomato disease are taken as inspecting and verifying objects for these methods. The multi-fault character is obvious in the systems that have typicality. The simulation results show that the proposed methods perform well and the proposed multi-fault (disease) diagnosis is effective, and the result corresponds to the diagnosis requirement. On the other hand, the methods can simulate human-thinking, distinguish mono-fault and multi-fault, and provide diagnosis strategy support.Finally, the range of applications, merits and demerits of the methods are discussed, all of the research in this paper is summarized. At the same time, the key points of multi-fault diagnosis researching in the future are proposed.更多还原