非线性主元分析故障检测和诊断方法及应用

本文针对间歇生产过程的特点，基于多方向主元分析方法(MPCA)和非线性理论，提 出了一种非线性多元统计分析方法——最小窗口方法，该方法突破了MPCA方法单模型、线性 化的建模方式，创新性地构造了适合间歇生产过程特点的多模型结构非线性建模方法，并侧 重于在线间歇过程性能监视和故障诊断的实时性，消除了预报未来测量值带来的误差，提高 了过程性能监视和故障诊断的准确率．本文详细地讨论了最小窗口PCA建模方法、原理、应 用实例．基于该方法设计的聚氯乙烯生产过程性能监视和故障诊断系统充分验证了该方法的 有效性．

STUDY OF A NONLINEAR PCA FAULT DETECTION AND DIAGNOSIS METHOD

In view of characteristics of batch process, this paper proposes a new real time and nonlinear minimum window principal component analysis (MWPCA) method based on multiway principal component analysis and nonlinear theory. MWPCA method breaks through linear MPCA modeling with single model structure and innovates in a nonlinear multi model structure for batch process modeling. The method emphasizes particularly on real time characteristic in on line batch process performance monitoring and eliminates error caused by predicting future measurements of process variables, increases the accuracy of process performance monitoring and fault diagnosis. MWPCA modeling procedures,principle and its application are discussed in detail. PVC process performance and fault diagnosis system based on MWPCA method verify the validity of the method.