数据挖掘技术及在机车故障诊断中的应用

【作者】 谢友辉；

【导师】 蒋新华；

【作者基本信息】 中南大学 ， 计算机应用技术， 2004， 硕士

【摘要】 数据挖掘研究如何从大量的数据中智能地、自动地提取出有价值的知识和信息，是当前相当活跃的研究领域。近年来，随着我国铁路信息化建设快速发展，知识的自动获取已经成为制约其进一步发展的“瓶颈”。因此，研究适用于机车故障诊断领域的数据挖掘，具有理论意义和重要的实用价值。 本文系统归纳总结了现有的数据挖掘技术，分析了粗糙集在数据挖掘应用中的特点。以SS8型电力机车主变流器为研究对象，通过对主变流器的工作原理的分析，建立了主变流器的仿真模型。以变流器的输出电压为故障信息分析对象，并仿真了变流器不同整流元件开路故障时的运行情况，构造了相应的输出电压波形。采用小波变换对电压波形进行能量分解，构造相应的特征向量。 由于特征向量的冗余，可能产生不可识别的故障。本文提出应用基于粗糙集的数据挖掘技术对变流器进行故障诊断，研究了故障特征的约简方法及诊断规则的获取方法，在保持识别能力的情况下，对规则进行修剪，得到了约简规则库。从仿真应用来看，所提出的应用方法具有较强的鲁棒性和泛化能力。 数据挖掘研究正处在发展阶段，数据挖掘本身以及其在机车故障中的应用还有许多问题值得探讨，本文的研究工作是一个尝试，相关工作还有待进一步深入。更多还原

【Abstract】 Data Mining, a new generation of tools and techniques for automatic and intelligent database analysis, is an active area with the promise for a high payoff in many business and scientific application. On the other hand, knowledge acquisition has been a bottleneck with the rapid development of railway information technology. To deal with this challenge, Data Mining technology is studied and applied to locomotive fault diagnose in this paper.The paper makes a summary of the current technology of Data Mining and analyses the feature of Rough Set in the application of Data Mining. Next, the SS8 Converter is listed as study target, through the analysis of the Converter’s work principle, the Converter’s simulated model is founded. According to Converter’s output voltage for analyzed target of fault information, the paper simulates the operation of the Converter different components in disconnection fault and structures relevant output waveforms. It adopts wavelet transform to make the power decomposition to waveforms and constructs relevant character vectors.Owing to the redundant character vector, it might cause some faults which cannot be recognized. Therefore, the paper adopts the technology based on Data Mining of Rough Set for Converter fault diagnose, study the method of fault character reduction and diagnose rule obtainer. The effective rule sets are acquired in terms of reduction of existing rules under remain identify ability conditions. When the results apply to the simulate data, it indicates all above methods have stronger robustness and general ability.Data Mining research are on the stage of developing. A lot of problems is worthy of discussing about Data Mining itself and its application in locomotive fault. The study of this paper is just to have a try, relevant work needs further study.更多还原