For modern large-scale complex dynamic processes

different measured variables have their own serial correlations and interactions among these variables show on different time points. A mutual information based dynamic fault detection method was proposed by an advantageously decentralized modeling strategy. After made multiple time-delayed observations on each variable

the relevant measurements for the variable were separated from all observations by utilizing mutual information and corresponding variable sub-blocks were created. This approach of variable grouping allowed each variable sub-block to capture sufficient information about its own self-and inter-correlations such that the dynamic characteristics of process data could be well analyzed. The principal component analysis (PCA) algorithm was employed to construct statistical modeling on each variable sub-block and a decentralized dynamic fault detection model for large-scale dynamic process. The feasibility and effectiveness of the proposed method on dynamic process modeling were validated by a case study of a chemical process.