Investigation of factors affecting coke strength after reaction (CSR) and developing a statistical model for CSR prediction

Ünsal, Barış
This study was aimed at investigating the coke strength after reaction (CSR) prediction before coke production by regression modelling. Initially, quality parameters of studied coals, namely Australian, American and Canadian coals, were categorized to understand fluctuation in the parameters. Parameters studied consist of proximate analysis, physical properties, rheological properties, ash chemistry, petrographical analysis and coke quality parameters of the coals. After understanding remarkable difference in coal quality parameters relative to origin, regression analysis was performed for the coals under study. Highly correlated parameters were detected by correlation analysis, performed via Excel and Minitab, considering both Pearson Correlation Coefficient and p - values. Devore states that two variables show strong relationship when correlation coefficient of them is above 0.8. Absolute values of correlation coefficients above 0.8 evaluated as highly correlated. Absolute values of correlation coefficients between 0.6 and 0.8 and p – values below 0.05 also evaluated as highly correlated. Then, best subset analysis was carried out by Minitab to indicate best alternative regression model. Decision of which parameters are included into model was given by evaluating R – square, R – square (adj) and R – square (pred) of best subset analysis model alternatives. For studied Australian, American and Canadian coals, CSR prediction models were developed individually. Categorization and origin base CSR prediction model development studies created the base of CSR prediction model for coal blends. Precision of the models controlled by mean hypothesis and whether residues of model are equal to zero or not was checked. 1 – sample t test, 2 – sample t test and one-way ANOVA test were used for mean hypothesis. In addition, Origin base CSR prediction models were comprised with formulas retrieved from literature. At the end of study, CSR prediction models were developed with 96.5 %, 93.41 %, 86.21 % and 80.99 % R – square for Australian, American, Canadian coals and coal blends respectively.