基于LNS-DEWKECA算法的多模态工业过程故障检测

受市场需求主导,工业过程需要在多种工作模态下切换,数据往往呈现多模态复杂分布特性,研究多模态的故障检测技术对于保障工业过程的安全运行具有重要意义.为此,提出一种基于局部近邻标准化(LNS)和方向熵加权核熵成分分析(DEWKECA)的故障检测算法.利用LNS实现多模态数据的标准化,相比于全局标准化, LNS可以有效消除多模态特性;考虑到故障样本与正常样本在变化趋势上的差异,定义样本变化方向的信息熵为方向熵,用来衡量样本变化方向的无序程度,从而利用DEWKECA实现数据降维,可以更有效提取数据变化方向特征;考虑到多模态数据往往服从非高斯分布,采用局部离群因子(LOF)算法建立监控统计量,根据核密度估计确定其控制限.最后,通过数值例子及TE过程仿真验证所提出算法的有效性.

Multimodal industrial process fault detection based on LNS-DEWKECA

Driven by the market demand, industrial processes need to switch between multiple modes, resulting in multimodal and complex distributed data. Therefore, the study of multimodal fault detection technology is of great significance to ensure safe operation of industrial processes. Aiming at the fact, a fault detection algorithm based on local neighborhood standardization(LNS) and directional entropy weighted kernel entropy component analysis (DEWKECA) is proposed. Firstly, multimodal data are normalized with LNS, which can eliminate multimodal characteristics efficiently when compared with global standardization. Considering the differences in changing trends between fault samples and normal samples, the information entropy of samples'' changing direction is defined as the directional entropy, which is used to describe the disorder degree of samples'' changing direction. So DEWKECA is used to realize dimensionality reduction, with which the characteristics of data''s changing direction can be extracted effectively. Given that multimodal data often obey non-Gaussian distribution, the local outlier factor(LOF) algorithm is adopted to establish monitoring statistics, whose control limit can be determined by kernel density estimation. Numerical example and the TE process simulation verify the effectiveness of the proposed algorithm.