【作者】 向长城；

【导师】 黄席樾；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 近年来,因关键设备故障而引起的灾难性事故时有发生,这些严重、灾难性事故的不断发生,迫使各国政府、社会以及科研人员高度重视对复杂系统故障预测和诊断方面的研究。由于缺乏对系统运行进行准确状态判断和健康分析,出于安全的考虑,对系统设备进行了大量不必要的维修,大大提高了运行成本。可随着现代工业及科学技术的迅速发展,工业设备日趋大型化、连续化、高速化和自动化,功能越来越多,结构也越来越复杂,传统的故障诊断技术已经不能适应复杂设备系统实际运行的需要。正是基于这个原因,近年来复杂系统的故障诊断和状态监测理论及应用研究引起了科研人员的极大关注,并取得了一些较好的成果。本论文基于可拓学关联函数的定量和定性反映系统状态的特点,融合神经网络、免疫系统的相关原理,以国家自然科学基金项目“基于相似性原理和免疫应答理论重构故障诊断系统”、湖北省教育厅科学研究项目“基于人工免疫网络的理论及应用研究”、湖北民族学院“基于可拓学的智能故障诊断研究”、国防预研项目“智能分布式指挥业务子系统决策支持平台”以及企业横向项目“励磁装置在线故障预测和故障诊断系统”和“航天产品电路漏电故障诊断系统”为背景,在克服传统算法的缺点上,通过对故障进行分类,设计了可拓免疫算法、可拓神经网络、可拓K近邻算法用于故障诊断、并利用可拓小波算法设计了故障诊断系统解决励磁系统断相故障、利用免疫可拓控制解决了起重机的状态监测。本文的主要研究成果包括:①以导师所承担的“励磁装置系统智能故障诊断和预测系统的开发”项目为研究背景,针对励磁系统的整流桥断相故障,结合信号分析方法傅立叶变换,小波包分析和可拓学的综合评价方法,开发设计了基于励磁系统的可拓综合故障诊断系统。②结合可拓学与人工神经网络,设计了可拓神经元和可拓神经网络,然后利用BP算法和遗传算法对可拓神经网络的结构和权值进行学习和优化,最后用故障案例实际验证。③利用可拓学与人工免疫算法相融合,开发可拓免疫算法用于智能故障诊断。利用人工免疫系统的免疫响应,自学习机制和免疫系统“自我-非我”识别理论,结合可拓学的关联函数,建立可拓距离函数,设计B细胞探测器物元和检测物元,然后利用免疫学习机制对B细胞探测器进行学习和训练,最后将训练好的B细胞探测器识别检测物元。④利用可拓学与K近邻算法相结合,用于数据分类和故障识别。首先对于分类的数据进行属性约简,然后利用可拓距离函数,度量训练样本之间的距离,这种距离优于其它距离的特点在于对于同一类的训练样本,进行统计分析,包括找出每个特征的最大值、最小值和均值,其次利用可拓距离度量每个测试样本与这一类样本之间的距离,最后利用K近邻的思想进行投票。通过UCI的两个基准数据集对该分类的正确率进行对比,最后通过故障样本实现了故障的分类。⑤状态监测是故障诊断过程中的重要组成部分,是预防和防止故障和事故发生的有效手段。通过对移动式起重机起吊物体过程的监测,找到预防起重机发生倾覆的办法。首先通过对起重机的力矩分析,找到影响该力矩的主要因素和危险因子函数,利用人工免疫网络算法对危险因子函数进行优化,找到影响每个影响因素的临界值,建立可拓关联函数的经典域。最后通过可拓集合和可拓变换,对起吊不同物重系统的健康度和危险因子进行评估,为预防系统发生倾覆提供有效的辅助支持。本论文的主要创新点:1)利用免疫系统的“自我-非我”识别原理和免疫克隆选择原理,结合可拓学的物元分析、可拓集合理论设计了可拓免疫算法,并用于汽轮机的故障诊断。2)提出了可拓距离的概念,并结合k近邻的思想,设计了可拓k近邻算法,用于数据分类和故障识别。3)结合免疫网络和可拓控制的思想,解决了起重机起吊物体过程中的状态监测问题。最后对全文的研究工作进行了总结,并指出了工作中进一步研究的方向。更多还原

【Abstract】 In recent years, disaster cases have frequently occurred in view of the breakdown of the key equipments, which compels the government of each country, society and researchers to highlight the study of the fault diagnosis and condition monitor in the complicated system. However, considering the matter of security, a great deal of unnecessary maintenance has been carried out on the premise of incorrect diagnosis and condition monitor for the operation system, which greatly increases operation cost. With the rapid development of modern industry and sci-technology, the industry equipments increasingly become large-scaled, continued, high-speeded and automated, which accordingly leads to more and more functions, and more and more complicated structures. Therefore, the traditional fault diagnosis method and technology far from meet the running demand of the complicated equipments system. For this reason, studies on the fault diagnosis and condition monitor theories and application of the complicated systems have aroused researcher’s great interest in recent years, furthermore, some great achievements have been abstained in this area.Based on the reflection system state features of qualitative and quantitative of extension relation function, with the combination of relative theories of neural network and immune system, With the background of the national natural science fund item--" Reconfiguration of Fault Diagnosis System Based on Immune Response Theory and Comparability Principle", the scientific research item--“artificial immune network theory and its application”carried by Hubei provincial Education Department,“Fault diagnosis based on fault diagnosis”and Enterprise horizontal item“On-line fault diagnosis and prediction system of excitation system”and“leakage electricity fault diagnosis of circuit system”, on the basis of overcoming the flaw of traditional algorithm, it is carried on the detailed analysis to extension immune algorithm, extension neural network, extension K nearest neighbor and extension comprehensive evaluation for fault diagnosis and extension control for condition monitor. The main research result include:①With the background of the item that my supervisor assumes,“the fault diagnosis and prevention system of excitation system development”, and the excitation system fault character-element as the study subject, the fault character of excitation signal is extracted by FFT and wavelet packets. With the utilization of the extension comprehensive evaluation method , the extension comprehensive fault diagnosis based on excitation system has be developed, and the results shows that the method can recognize fault signal.②The extension neural network and extension neural element have been designed on the premise of extension and artificial neural network. And then, through BP algorithm and genetic training algorithm, the structure of extension neural network and power are trained and optimized, finally, being tested through the rotary fault.③The Extension Immune Fault Diagnosis Algorithm(EIFDA), which is proposed through extension and artificial immune system, can be used for artificial fault diagnosis. The immune response mechanics and“self and non-self”recognition principle are utilized to identify fault, and extension distance function is designed. Based on above principle, have been established B-cell detector matter-element and test matter-element. The B-cell detector is trained through the immune learning and the fault sample data. The different mature B-cell detector is used to recognize the test ample. The result shows the EIFDA is an effective algorithm for fault diagnosis.④In this part, a new Extension K Nearest Neighbor (EKNN) is proposed for data classification problem and fault diagnosis. Firstly, different attribute numbers impacts the classification correct rate, the data attribute reduction is applied by genetic algorithm. Secondly, the EKNN is described by K Nearest Neighbor(KNN) and extension distance. For each reduction attribute of training samples, the maximum、minimum and mean is computed. The extension distance is measured the test sample and the training samples .The K nearest extension distance allocates the different classes. Finally, two data of UCI and a fault diagnosis sample is used by EKNN, the results shows the EKNN classification correct rate is better than other traditional algorithm.⑤Condition monitor, one of significant parts of the fault diagnosis, is an effective measure to prevent the fault occurrence. In this part, the aim is to find out the method to prevent the traveling crane from tilting through careful monitoring the process of which the traveling crane works. Firstly, the aim is to work out the main factor and danger function index through the analysis of the moment of the crane. Secondly, the impact factor bound value is optimized by immune network algorithm, which helps to work out the classical zone of extension conjunction function that influences each influential factor. Finally, the evaluation for the health evaluation system under the condition of the systems with different heavy goods lifting and danger function, based on extension set and extension transformation, provides an effective assistant support for preventing system from tilting.To sum up ,the creativity of this thesis lies in the following aspects:, first, the extension immune algorithm, has been successfully designed and can be used for steam turbine fault diagnosis, on the basis of utilization of“self and non-self recognition theory in immune system and immune clone selection theory, with the combination of extension matter-element analysis and extension set; second, the concept of extension distance has been put forward and on the basis of it the extension K nearest neighbor has been designed which can be used for data classification and fault diagnosis; third, the problem of condition monitor for the crane in the process of goods lifting has been settled on the basis of immune network and extension control.In the final part, the summary to the research work of the full text is given, furthermore, the direction of further study is pointed out that theories and application in fault diagnosis and condition monitor of the complicated system.更多还原