基于局部密度估计的多模态过程故障检测

Multimode process monitoring based on local density estimation

以市场需求为导向的现代工业过程的生产条件要根据市场的需求不断做出调整,因此实际工业过程中存在多种工况的复杂情况,而过程的数据将不再完全服从高斯分布,其均值与协方差结构往往随着工况的切换而发生较大变化,为了能及时检测此类生产过程中的故障,提出一种新的基于带宽可变的局部密度估计的过程在线监控策略。首先利用局部投影保留(locality preserving projection,LPP)将高维数据投影到低维子空间中,充分地保留数据的局部结构;然后通过带宽可变的非参数密度核函数来进行局部密度估计,并采用局部密度因子(local density factor,LDF)的思想构造监控统计量,进而对工业过程故障进行在线检测;最后通过仿真研究,结果表明所提方法能够有效地应用于多模态过程的故障检测。

Multiple operating modes and complex data distributions are common in modern industrial processes, so the process variable does not completely obey Gaussian distribution, and its mean and covariance structure tend to change with the switching of working conditions. In order to detect various faults in such production process timely, a novel online monitoring strategy based on locality preserving projection （LPP） and local density estimation with variable bandwidth is proposed. Firstly, LPP is introduced for feature extraction and dimension reduction, and then a variable bandwidth kernel density estimate function is adopted to compute every sample’s local density factor, which is used as a monitoring statistic value. Finally, the results of numerical examples and TE chemical process simulation experiment validate the effectiveness and utility of the proposed method for the fault detection of multi-mode processes.