A statistical analysis of data from the production line at the Munksund paper
mill
Halfdan Grage
Centre for Mathematical Sciences
Mathematical Statistics
Lund Institute of Technology,
Lund University,
2004
ISSN 14039338

Abstract:

A multivariate statistical analysis of paper strength and production process
data from the SCA Munksund paper mill is presented. The analysis is divided
into five parts: (i) paper quality variables, (ii) process variables from
the paper machine, (iii) process variables from digester and refiner, (iv)
a combination of all process variables, and (v) a further analysis of the
zstrength.
Analysing the correlations between the paper strength variables, two groups
of variables could be discerned: one group comprising the variables tensile
strength in MD, tensile stiffness in MD, elongation in CD, and bursting strength,
and another group containing the variables SCT in CD, tensile strength in
CD, and tensile stiffness in CD. The strongest correlations between paper
machine process variables and paper strength variables were found for the
strength variables in the first of the above mentioned groups. The same strength
variables were negatively correlated with the wet end process variables
conductivity, base line paper machine flow, and the amount of return fibre,
while they were positively correlated with the concentration at the base
line refiner. Combining wet end and paper machine process variables in a
regression analysis for the paper strength variables yields a limited increase
in the amount of variance explained, compared with the case where only wet
end process variables are used as predictors, due to the correlations between
the process variables. A more detailed study of the time series of zstrength
measurements showed that

this was nonstationary, which makes it hard to fit a single overall linear
regression model which will give good predictions over the whole year or
even from one production run to another.




