Hierarchical multivariate analysis of cockle phenomena

Abstract

The phenomena called cockle are small wrinkles on the paper surface that appear during paper production. This condition poses significant economic and operability problems in the production of magazine paper, as it deteriorates the printabilty of the paper. There are many and varied sources that can lead to cockle, and their detection is often very complicated. In this work a multivariate hierarchical approach is proposed to analyze the cause of cockle. The hierarchy has two levels, the first of which is a three-way decomposition and analysis of the data collected from sections of a paper machine. The second level is a two-way decomposition and analysis between the combined loadings from the three-way decomposition and the measured cockle data. The results show that this approach is capable of identifying the important process sections and process variables, in spite of the large dimensionality of the problem. Data analyzed from two real industrial paper machines, involving several grades of paper, are used to demonstrate the proposed hierarchical approach.

Application of steady-state and dynamic modeling for the prediction of the BOD of an aerated lagoon at a pulp and paper mill

“…Outliers were identified graphically by plotting principal components analysis (PCA) and PLS scores [13,14]; newer robust multivariate methods of outlier identification are proposed by Hoo et al [15]. The obtained results were analyzed together with the IPB process specialists who suggested that only the pH outlier data collected on 6 April 1997 should not be used to construct the models.…”

“…Only the variable color does not seem to be important to the PLS model. For problems with larger dimensionality, the use of specific approaches for identification of variable importance [14] is essential.…”

Tensor-based anomaly detection: An interdisciplinary survey

“…For semi-supervised and unsupervised techniques, true and false positives (or false alarms) are also assessed [26,43,65,126] . For regression based tensor models and tensor forecasting methods, prediction error metrics such as root mean square error (RMSE) or mean absolute error (MAE) is normally used [50,59,71,90,92,124,136,154] . Detection delay has also been exploited in some works [40] .…”

Application of steady-state and dynamic modeling for the prediction of the BOD of an aerated lagoon at a pulp and paper millPart I. Linear approaches

“…Outliers were identified graphically by plotting principal components analysis (PCA) and PLS scores [13,14]; newer robust multivariate methods of outlier identification are proposed by Hoo et al [15]. The obtained results were analyzed together with the IPB process specialists who suggested that only the pH outlier data collected on 6 April 1997 should not be used to construct the models.…”