中文摘要：

现有的多尺度主元分析方法为监测具有多尺度特性的工业过程提供了一种有效途径,但该方法还存在以下两个问题：一是采用了重构步骤使得需要建立的监测模型数大大增加;二是采用Haar小波进行小波变换,而Haar小波不连续从而对信号特征的刻画能力比较弱,为此,本文提出了根据故障尺度特征的分布特点修改原有的多尺度主元分析的框架,去除了重构步骤并具体给出了突变故障和振荡故障的定位和跟踪方法,还提出了采用sym小波进行多尺度分析并解决了边界效应的处理和信号对齐的计算等问题,在一个标准的CSTR仿真过程中验证了所提方法的有效性。

英文摘要：

The existing multiscale principal component analysis method provides an effective way to monitor industrial processes with multiscale features due to the influence of events at different time-frequency values, but there are two main shortcomings with respect to this method： a reconstruction step is needed which results in a great number of monitoring models to be constructed; and Haar wavelet is used to do a wavelet transform for multiscale analysis, but it is not continuous and therefore not good at approximating practical signal. Hence, a modified multiscale monitoring method was proposed, which not only eliminated the reconstruction step based on the analysis of scale feature of the fault but also replaced Haar wavelet with sym wavelet, which was continuous and had a higher order vanishing moment and thus was more suitable to extract multiscale features. In particular, the proposed method gave an alternative multiscale way to determine the location and duration of two typical faults ： step fault and oscillating fault, and solve the problem of boundary effect and signal alignment accompanying the introduction of sym wavelet. Finally, the effectiveness of the proposed methods was verified by simulated experiments on a standard CSTR problem.