- Title
- Assessing coal properties and their effects on coking performance: a data mining approach
- Creator
- North, Lauren
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2020
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Within the cokemaking industry, the ability to accurately predict the quality of coke produced from a variety of global coal basins is limited. Since the most recent review of coke quality prediction models, completed by Dı́ez et al. (2002) on a selection of key models within the literature, very little improvement in prediction ability has been shown. In particular, the existing coke quality prediction models are deficient in their ability to link and expand upon reported fundamental coal behaviour. Where emerging data mining techniques have been applied for model development, often a clearly defined and technically sound analytical process has not been described. In this context, data mining as one step in the knowledge discovery process, has presented a unique opportunity to provide further insight into the behaviour of coking coals, and in particular, allowed for integration and extension of fundamental coking behaviour. Based on the conclusions of an extensive literature review considering both coking behaviour, and analysis of the methods of prediction of coke quality, a sub-model approach to prediction was developed. This sub-model approach addresses the some of the fundamental processes occurring within the formation of coke from the parent coals, whilst integrating data mining techniques. The following knowledge gaps were explored, with clearly defined frameworks for prediction developed: Vitrinite reflectance is perceived as an important parameter in coke quality, however in coal blending, utilising the average value is unsuitable as it does not reflect how two coals of vastly different reflectance have been blended. Classification of the resulting distributions allows further insight into these blending decisions. Traditional prediction models consider fusibility of coal sub-components as one of the main factors in determining coke quality. However, the assumed proportion of fusing and non-fusing components is unreliable for Australian and other coals. Hence, accurate prediction of these fusing components for each respective coal ought to improve prediction of coke quality. Data quality issues associated with the comparatively small and biased data set were also explored. This section discusses how some of these issues can be addressed by using a modified data oversampling technique based on the Synthetic Minority Oversampling Technique (SMOTE). Coal ash chemistry is commonly used in the prediction of coke reactivity. However, the use of ash chemistry may be misleading as it does not directly reflect the mineral forms present in the coal which may have different influences on reactivity. Whilst basicity index, derived from coal ash chemistry, is often applied in coke quality predictions, questions are raised over the relative importance of different minerals on coke reactivity, and whether the index is misleading. Using the approximated coal mineral composition, an improved understanding of coal reactivity behaviour is developed. Results show statistically significant improvement in predictive accuracy with the use of the vitrinite reflectance distribution terms, as well as the use of coal mineral composition. The development and integration of these models in a clearly defined predictive framework presents a significant improvement to many traditional predictive models. Further, consistency with experimental studies was observed, with several areas for future work identified. This work has significant implications on not only the methods of prediction of coke quality, but also on the confirmation and integration of experimental findings.
- Subject
- cokemaking industry; coal basins; coke reactivity; data mining; predictive accuracy
- Identifier
- http://hdl.handle.net/1959.13/1425636
- Identifier
- uon:38285
- Rights
- Copyright 2020 Lauren North
- Language
- eng
- Full Text
- Hits: 420
- Visitors: 513
- Downloads: 128
Thumbnail | File | Description | Size | Format | |||
---|---|---|---|---|---|---|---|
View Details Download | ATTACHMENT01 | Thesis | 64 MB | Adobe Acrobat PDF | View Details Download | ||
View Details Download | ATTACHMENT02 | Abstract | 501 KB | Adobe Acrobat PDF | View Details Download |