Abstract: Principal component analysis (PCA) was an important and most widely used data whitening approach in process monitoring based on independent component analysis (ICA) due to its effectivity in reducing dimensions of objects. PCA model was generated based on sample data of normal process, and the first several PCs which contain the most variance information of normal process were employed for ICA and process noise was eliminated. In PCA model, T2 statistic of each principal component has the property that it can measure variation along direction of the component. By researching the T2 statistic of sample data of fault process, it is found that information of some faults are mostly reflected on the components corresponding to smaller variance contribution, which are regarded as process noise and eliminated, and thus missed detection problem is happened. At last, the effect of PCA whitening on chemical process monitoring based on PCA-ICA is illustrated through simulations of both a simple process and TE process, and the results prove the proposed opinion.

Key words: Principal component analysis, Data whitening, Independent component analysis, Process monitoring

摘要： 主元分析是基于独立元分析过程监控中一种重要而且常用的白化方法，可以有效地降低监控对象的维数。其基于正常样本数据，根据主元方差贡献率选取主元，保留正常样本中的大部分方差信息，消除噪声。在ＰＣＡ模型中，每个主元的Ｔ２ 统计量表征着样本数据沿该主元方向的变异程度。通过对故障样本数据每个主元的Ｔ２ 统计量分析，发现某些故障信息投影在方差较小且被舍弃的主元上，从而造成故障信息的损失，进而影响了ＩＣＡ的监控性能，造成故障的漏检和故障源的误识别。最后，采用一个简易系统和ＴＥ过程，验证了ＰＣＡ 白化过程对ＩＣＡ监控性能的影响。

关键词: 主元分析 , 数据白化 , 独立元分析 , 过程监控