基于局部临近标准化的FD-KNN故障检测

FD-KNN Fault Detection Based on Local Nearest Neighborhood Standardization

针对方差相差大的多模态故障诊断数据问题,提出一种基于局部临近标准化(local nearest neighborhood standardization,LNNS)的k近邻故障检测方法(fault detection-k-nearest neighbor rule,FD-KNN)。首先,计算每个样本的局部近邻,采用近邻特征实现标准化,克服传统标准化方法 Z-score将多模态数据看成一个整体而使数据不准确问题;其次,计算每个样本间距离,建立局部临近标准化距离模型,通过临近距离确定控制限。最后,在半导体生产过程中进行仿真应用研究,通过实验结果的比较与分析表明了所提方法的有效性。

For large variance of multi-mode fault diagnosis data, this paper presents the ^-nearest neighbor fault de-tection method （FD-KNN） based on local nearest neighborhood standardization （LNNS） . Firstly, the local nearest neighbor of each sample was calculated and the local nearest neighborhood feature was used to achievization so as to overcome the inaccurate data produced by the traditional standard method of Z-score because it took the multi-mode data as a whole. Secondly, the distance between samples was calculated and a hood standardization distance model was established to determine the control limits based on local distansimulation study was conducted in the process of semiconductor manufacturing and the effectimethod was verified by analyzing comparing the experimental results with those of