Double-weighted neighborhood standardization method with applications to multimode-process fault detection

As modern industrial processes often have multiple production modes, multimode-process monitoring has become an important issue. In multimode processes, the operating condition may often switch among different modes. As a result, popular process monitoring methods such as principal component analysis (PCA) and partial least squares (PLS) method should not be directly applied because they are based on a fundamental assumption that the process only has one stable operating condition. In this paper, a novel multimode-process data-standardization approach called double-weighted neighborhood standardization (DWNS) is proposed to solve the problem of multimode characteristics. This approach can transform multimode data into approximately single-mode data, which follow a Gaussian distribution. By analyzing a concrete example, this study indicates that the DWNS strategy is effective for multimode data preprocessing. Moreover, a novel fault detection method called DWNS-PCA is proposed for multimode processes. Finally, a numerical example and the penicillin fermentation process are used to test the validity and effectiveness of the DWNS-PCA. The results demonstrate that the proposed data-standardization method is suitable for multimode data, and the DWNS-PCA process monitoring method is effective for detecting faults in multimode processes.