Developing A DPCA-Dr Based Glrfault Detection Method For Autocorrelatedhigh-Dimensional Process
As the development of Industry 4.0, the field of high-dimensional fault detection plays an important role in ensuring the online production quality. The Principal Component Analysis (PCA)is a widely used high-dimensional process monitoring method. However, the conventional PCA fails to monitor the autocorrelated processes. Hence, the Dynamic PCA (DPCA) was developed in an attempt to monitor the autocorrelated processes. Researchers found that the DPCA’s monitoring statistics and Q still exhibit autocorrelation which violates the prerequisite of PCA implementation. Therefore, DPCA with Decorrelated Residuals (DPCA-DR) was proposed to time-decorrelate the monitoring statistics. Even though the DPCA-DR can perform well for autocorrelated processes monitoring, it is insensitive to detect the small process changes. In this study, the DPCA-DR based Generalized Likelihood Ratio (DPCA-DR-GLR) charting statistic will be proposed. The proposed method has advantages of 1) detecting a wide range of process changes, 2) estimating the change points, 3) needless prior parameters to be specified by practitioner and 4) only one chart to be plotted. The efficiency of the proposed method will be verified via a simulated autocorrelated process. Results demonstrated the proposed possesses superior performance than traditional high-dimensional monitoring methods.
Keywords - PCA, DPCA, DPCA-DR, GLR, fault detection.