中文摘要：

针对动态非平稳过程数据的时变性和多尺度性导致故障诊断准确率下降及故障准确定位难以实现的不足，引入滑动窗口多尺度主元分析，通过小波阈值消噪解决统计模型偏离与数据相关性降低之间的矛盾，并在各个尺度上利用滑动窗口主元分析实现模型更新，然后借助三维贡献图描述反映过程行为变化的各独立过程变量对统计过程的贡献程度，进而对故障准确定位，最后给出诊断准确性的定量评价机制。在对6135D型柴油机进行数值实验中，并通过与传统的多尺度主元分析及自适应多向主元分析比较，实验结果从故障的漏报率、误报率及诊断准确率三方面表明新方法能更好地实现传感器故障诊断。

英文摘要：

To track the non-stationary dynamics of the process which contains time-varying and multi-scale data, an online moving window multi-scale principal component analysis（MW-MSPCA） data-driven-based fault diagnosis method is proposed. In this data-driven diagnosis technique, wavelet threshold denoising is used to solve the conflict between the statistical model deviation and data correlation decreasing. The statistical models are updated by using moving window principal component analysis in various scales. The contribution of individual process variable to the process behavior change is illustrated in a 3-dimensional contribution chart. A quantitative evaluation mechanism is also given to evaluate the diagonising accuracy. The numerical experimental results for 6135D diesel demonstrate that the proposed method can diagnose sensor fault better in terms of false rejection, false alarm and diagnosing accuracy for fault diagnosis upon comparing with conventional multi-scale principal component analysis （MSPCA） and adaptive multi-way principal component analysis（AMPCA） modeling.