Data compression and soft sensors in the pulp and paper industry

Thomas A. Runkler, Erwin Gerstorfer, Martin Schlang, Erwin Jiinnemann, Klaus Villforth

Abstract: Two key problems in industrial plant optimization are the compression of data from the automation system and the estimation of values which are not directly available. Clustering can be used to determine technologically meaningful operating points from data sets which serve as compressed archive data. Block selection techniques yield a speedup that makes this method feasible for industrial applications. Clustering can also be used to generate nonlinear models from sensor and laboratory data. These models are used as soft sensors which give good online estimations of variables which can only be measured offline in the laboratory. Both methods, data compression and soft sensor, are applied to the optimization of the deinking process in recovered paper processing in the paper industry.