基于多块核主元分析和概率符号有向图的故障诊断方法研究

发布时间：2018-04-03 03:22

本文选题：符号有向图　切入点：多块核主元分析　出处：《中南大学》2013年硕士论文

【摘要】：摘要：故障诊断是工业生产特别是流程工业中的一个重要问题,在过去的几十年中有大量致力于这方面的研究。基于符号有向图(Signed Directed Graph, SDG)模型的故障诊断方法,由于具有表达复杂因果关系和包容大规模潜在信息的能力,完备性好、适应性强,同时可提供故障传播路径和演变解释,得到了学者们的广泛关注。然而,基于SDG模型的故障诊断方法作为纯定性方法存在分辨率低且所建SDG模型精确度、可靠性不高等问题。针对SDG存在的问题,将传递熵、多块核主元分析(Multiblock Kernel Principal Component Analysis, MBKPCA)、概率论等定量处理方法与SDG相结合,开展基于SDG的故障诊断方法研究,具有科学意义和应用价值。论文主要研究工作及创新成果如下： (1)针对传统SDG建模方法中过程知识不完备、错误以及精确数学模型缺失的不足,引入传递熵进行SDG建模。考虑到系统中各变量之间的传递熵既能量化变量之间的依赖性,又能测量变量之间的方向性,充分利用过程历史数据,基于传递熵构建变量节点链模型,实现SDG建模,从而提高SDG建模的准确性和可靠性。 (2)针对大规模工业过程变量之间关系复杂,故障难确定难诊断的特点,提出一种基于MBKPCA口SDG的故障诊断方法。首先,提出基于SDG和优先级的分块策略,以强连通元SCC为最高优先级、多入＼出度节点群为次高优先级、节点链为最低优先级对过程进行分块；在此基础上,采用MBKPCA进行过程监控,对于检测到的故障,先确定故障发生在哪一个数据块,再触发SDG在故障块内完成故障定位。所提方法克服了MBKPCA故障隔离不完全和SDG推理过程中组合爆炸的缺点,可提高复杂工业过程故障诊断的准确度与速度。 (3)对于存在正反馈回路而无法使用上述方法实现故障完全隔离的特殊情况,进一步提出改进的概率符号有向图(Probabilistic Signed Digraph, PSDG)方法,提出新的环状结构打开方法并确定打开后的支路概率,根据计算出的后验概率对可能的故障根源进行排序,以此概率排序为依据隔离故障,确保SDG故障诊断流程的完整性。基于Tennessee Eastman过程的仿真研究验证了所提故障诊断方法的有效性。论文共有图37幅,表7个,参考文献80篇。
[Abstract]:Absrtact: fault diagnosis is an important problem in industrial production, especially in process industry.The method of fault diagnosis based on symbolic directed graph signed Directed Graph (SDGs) model, because of its ability to express complex causality and contain large scale potential information, is good in completeness and adaptability, and can also provide the path of fault propagation and explanation of evolution.It has received extensive attention from scholars.However, as a pure qualitative method, the fault diagnosis method based on SDG model has some problems, such as low resolution, accuracy and reliability of the established SDG model.Aiming at the problems existing in SDG, combining the quantitative processing methods such as transfer entropy, multiblock Kernel Principal Component analysis, MBKPCAA and probability theory with SDG, it is of scientific significance and application value to develop the research of SDG based fault diagnosis method.The main research work and innovative results are as follows:1) aiming at the defects of incomplete process knowledge, errors and lack of accurate mathematical models in traditional SDG modeling methods, transfer entropy is introduced to model SDG.Considering that the transfer entropy of each variable in the system can not only quantify the dependence between variables, but also measure the directivity of variables, make full use of the process history data, construct the model of the node chain of variables based on the transfer entropy, and realize the SDG modeling.In order to improve the accuracy and reliability of SDG modeling.A fault diagnosis method based on MBKPCA port SDG is proposed to overcome the complex relationship between variables in large-scale industrial process and the difficulty of fault diagnosis.Firstly, a block strategy based on SDG and priority is proposed, in which the strongly connected element SCC is the highest priority, the multi-input / output node group is the next high priority, and the node chain is the lowest priority to block the process.The MBKPCA is used to monitor the process. For the detected faults, it is determined which data block the fault occurs, and then the SDG is triggered to complete the fault location in the fault block.The proposed method overcomes the shortcomings of incomplete MBKPCA fault isolation and combined explosion in SDG reasoning process and can improve the accuracy and speed of fault diagnosis in complex industrial processes.3) for the special case where there is a positive feedback loop and the above method can not be used to achieve complete fault isolation, an improved probabilistic Signed graph (PSDG) method is proposed.A new opening method of annular structure is proposed and the open branch probability is determined. According to the calculated posteriori probability, the possible fault source is sorted in order to isolate the faults according to the probabilistic ranking to ensure the integrity of the SDG fault diagnosis process.Simulation results based on Tennessee Eastman process verify the effectiveness of the proposed fault diagnosis method.There are 37 pictures, 7 tables and 80 references.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TH165.3

【参考文献】