摘要： 鉴于传统的多向主元分析(MPCA)难以保证在线状态监测和故障诊断的实时性，提出了一种基于特征子空间的滑动窗主元分析(CSMWPCA)故障监测与诊断方法。在实时故障监测与诊断时，该方法采用适当大小的滑动窗逐步更新当前子数据空间，对当前子数据空间故障的识别通过依次计算其与基底库中各故障的匹配度来进行，克服了传统的MPCA 不能处理非线性过程和实时性问题，与一种新的移动窗多向主元分析(MWMPCA)方法相比，CSMWPCA 方法能更有效地识别故障发生的原因。

关键词: 主元分析；特征子空间距离；滑动窗口；批过程；故障诊断

Abstract: A characteristic subspace moving window principal component analysis (CSMWPCA) for on-line batch process monitoring and fault detection is proposed. Using proper moving window to update current data subspace and calculating matching degree between the current data subspace and each fault belonged to fundus warehouse step by step, the approach recognises the current data subspace fault and emphasizes particularly on-line process performance monitoring and exactly fault detecting which results in extraordinary behavior of batch processes.Compared with a new moving window multiway principal component analysis (MWMPCA), the characteristic subspace moving window principal component analysis (CSMWPCA) can more effectively recognize fault reason.

Key words: Principal component analysis; Characteristic subspace distance; Moving window; Batch process; Fault detection