【作者】 王（王莹）；

【导师】 张颖伟；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 连续采煤机是煤矿中的重要设备,如果连续采煤机因为发生故障而停机,会导致整个煤矿系统瘫痪。因此,对连续采煤机进行故障检测具有重要的现实意义。本文提出了一种基于改进的核独立元分析(Kernel Independent Component Analysis,简称KICA)的故障检测方法,并针对连续采煤机截割部减速器进行故障检测。KICA结合了核主元分析(Kernel Principle Component Analysis,简称KPCA)和独立元分析(Independent Component Analysis,简称ICA)的优点,是在线故障检测的一种非线性方法。在KICA方法中,数据映射到特征空间后变得线性冗余；当引入核技巧时,输入空间线性相关的数据映射到特征空间会产生误差；另外,在KPCA训练过程中核矩阵大小是样本个数的平方,计算量较大。针对上述问题,本文提出了相似性分析的解决方法,对KICA进行了改进,即在使用KICA算法之前先在输入空间和特征空间对数据进行相似性分析,去除掉相似性较强的数据。此方法既降低了计算量,又减少了引入核技巧时带来的误差。利用改进的KICA我们能够从连续采煤机的数据中提取具有代表性的特征数据,根据其提取的特征数据计算故障检测统计量,能够更好的反映当时连续采煤机采煤过程的运行状况。本文采用Hotelling T2和平方预测误差(SPE)统计量进行故障检测,由于数据从输入空间映射到特征空间时,原始的SPE统计量计算公式不再适用,因而针对改进的KICA建立了一种新的SPE统计量计算公式。最后本文将此方法应用到了连续采煤机的截割部减速器故障检测当中,并通过Matlab进行了仿真研究,实验结果表明改进的KICA有效地捕获了变量中的非线性动态特征,并成功检测到故障的发生。更多还原

【Abstract】 The continuous miner is an important equipment in the excavate coal industry. It may make the coal-mining system broken-down, if continuous miner stop running cause of fault.The importance of the continuous miner fault diagnosis is increasingly prominent. In this paper, the fault detection method based on improved kernel independent component analysis (KICA) is developed. The proposed approach is applied to cutting unit reducer of continuous miner for fault detection.KICA is unsupervised and combines the advantages of KPCA and ICA to develop a nonlinear approach to detect fault online. Because the data mapped into feature space become linearly redundant, linear relation data introduce error while the kernel trick is used and the size of the kernel matrix are the square of the number of samples in the training process of KPCA, similarity analysis is developed to improve KICA algorithm, i.e., observation data is deal with using similarity analysis in input space and feature space before KICA algorithm. This method not only decreases the computation load, but also reduces the error while the kernel trick is used.We can extract the dominant feature data from the data of continuous miner by the KICA method. The statistics, calculated based on the feature data, can be better to describe the mining status of the continuous miner. This paper makes use of the Hotelling T2 and SPE statistics for fault detection. The original formula for calculating the SPE statistics is not available any more, because the data is mapped from input space to feature space. To solve this problem, a new formula for KICA is established.This method is applied to the fault detection of the reducer of continuous miner in this paper. The simulation result shows that the KICA method can capture the nonlinear dynamic features effectively, and detects the fault successfully.更多还原