Product Quality Assessment and Controller Performance Monitoring in Paper Machines, 1998 Engineering Conference Proceedings
Principal component analysis (PCA) is used to separate full sheet profiles into spatiotemporally correlated and independently random components. Dominant profile features are extracted from measured stochastic data using Karhunen-Loeve (KL) expansion. The method is statistically optimal and it adaptively reconstmcts the best orthogonal basis functions for each different profile. KL approximation is shown to work equally on total sheet and traditional residual profiles (which usually retain correlated data after MD and CD average profile separations). The additional statistical knowledge gained through the new approach is used to improve sheet quality assessment and to define controller and process performance metrics. It is shown that KL approximation is also an efficient data compression method.
Product Quality Assessment and Controller Performance Monitoring in Paper Machines, 1998 Engineering