Multi-PCA models for process monitoring and fault diagnosis

Multivariate statistical approaches have been proved effective for reducing the dimension of highly correlated process variables and subsequently simplifying the tasks of process monitoring and fault diagnosis. However, for the process with distinctive stages, a single statistical model is not sufficient or even incapable to map the substantive process information. In this paper, multi-PCA models are proposed for promptly detecting faults and improving the exactness of the diagnosis as well. The effectiveness of the approach is demonstrated on a complicated fermentation process.