Development of a predictive model for the moisture content in a paper mill

Dijkstra, B.K. (2011) Development of a predictive model for the moisture content in a paper mill. Master's Thesis / Essay, Chemistry.

Preview	Text Development_of_a_predictive_mo_1.pdf - Published Version Download (2MB) | Preview
	Text AkkoordDijkstra.pdf - Other Restricted to Registered users only Download (14kB)

Abstract

In the production of paper, one important paper quality parameter is the moisture content. However controlling paper quality is a difficult task as the measurement of this quality is affected by a large dead time in the production process. In order to overcome this problem, one possibility is to develop a predictive model based on the key variables affecting the paper quality, that are measured more frequently. In this way it should be possible that the paper quality parameter can be determined at an early stage of the production process. This thesis describes the development of such a predictive model for the Sappi paper mill in Nijmegen. Analysis of the raw process data with the use of principal component analysis yielded that the process data was arranged in clusters of consecutive data. The cause of this arrangement was found in the variation of the average values of the process input variables. Although many modelling techniques are available, in this thesis a linear regression technique (PLS) as well as a non-linear technique (Robust LSSVM) was used for the development of a predictive model. It was found that both techniques showed comparable results for the prediction of the paper quality, therefore for the predictive models, PLS was preferred because of its simplicity. The initial developed PLS models in combination with a bias update showed promising results regarding the goodness of fit. However the models used a too small range of variables close to the measurement of the paper quality, therefore these models are not real predictive models. However these models can be used as a substitute measurement of the paper quality in case the real measurement fails. The final PLS model is based on manually controllable variables and important process values, in combination with an online update of the bias term in the PLS model. Due to the nature of these variables, this model is more likely to be used for controlling the paper quality. The final model showed a good fit both with and without the online update incorporated. One remark has to be made however. The developed model is only tested for the production of paper with a gram weight of 90 g/m2. As the paper mill produces a wider range of gram weights, it is required to perform the analysis presented in this thesis again for the other paper gram weights.

Item Type:	Thesis (Master's Thesis / Essay)
Degree programme:	Chemistry
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	15 Feb 2018 07:46
Last Modified:	15 Feb 2018 07:46
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/9771

Actions (login required)

View Item