Modelling the effect of production process parameters on the dispersion capabilities of lignosulphonates

Thumbnail Image

Date

2022

Authors

Dhanpat, Jennica

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

The Sappi Tugela Mill produces lignosulphonate as a by-product of the Neutral Sulphitem Semi-Chemical (NSSC) pulping process. The product is primarily used as a dispersant in the concrete and cement admixture markets, where it is blended with other plasticisers such as polycarboxylate ethers. Attempts to predict the lignosulphonate dispersion characteristics and align them with market demands have been a technical challenge for researchers in this field of industry. Thus, this study aimed to understand and model the effects of process parameters on lignosulphonate dispersion capabilities. Currently, three offline methods are used to assess the dispersion performance of the lignosulphonate produced: the insoluble content, the dispersion index, and the concrete slump test. As a result, predictive models based on process and product knowledge, as well as data analysis using various regression modelling methods, were used to estimate these measurements. RapidMiner and Microsoft Excel were used to develop these models. The concrete slump data comprised five measurements taken at 15-minute intervals, beginning with an initial value. When the training data were normalised by the initial value, the trends of the data were reasonably linear, implying that all the slump data can be fitted with two parameters, the initial value and time. A general model for predicting concrete slump behaviour was found, in which the slump data was defined by a simple quadratic function and a Neural Net model was developed, using process parameters to predict the initial concrete slump values. Using production data, Random Forest models were developed to predict the insoluble content and dispersion index values. The developed models' results were compared to actual laboratory values, and a simple adaptive approach, of bias updating, was used to improve the models' fit to the data. This work developed a useful system for predicting the lignosulphonate dispersion performance, thus reducing the reliance on laboratory results, and allowing for almost immediate changes to be made to the dispersing agent production process. The modelling approach used in this study proved successful. The models' predicted values generalized to a reasonable degree to the test sets and captured the trends of the actual values, but the models made significant prediction errors. The reasons for this performance are presented, and this includes the optimal hyperparameters obtained and the conditions under which these tests are conducted. Improving model performance through bias updating resulted in only minor differences between model predictions and actual laboratory values. Therefore, signifying quality models were developed for a lignosulphonate prediction system or implementation into a suitable control strategy.

Description

A dissertation submitted in fulfilment of the requirements for the degree of Master of Science in Engineering to the Faculty of Engineering and Built Environment, School of Chemical and Metallurgical Engineering University of Witwatersrand, Johannesburg, 2022

Keywords

Citation

Collections

Endorsement

Review

Supplemented By

Referenced By