Multivariate statistical assessment of coal properties

Abstract

A set of 42 coal samples consisting of coal blends prepared for coking (subset A-blends) and lump coal from coal seams (subset B-single coals) was subjected to multicomponent statistical analysis. For these samples, the study determined their structural properties (total intrusion volume TIV, total pore area TPA, bulk density BD, average pore diameter APD, and porosity PS), proximate characteristics (moisture Wa, ash content Ad and volatile matter Vdaf), ultimate characteristics (total sulfur content Sd and carbon content Cd), coal maceral characteristics (reflectance of vitrinite Rr, vitrinite Vitr, inertinite Inert and liptinite Lipt) and coking properties (contraction a, dilation b and swelling index SI). Using factor analysis (FA), 3 factors were separated. These include the most important coal characteristics with significant mutual correlations. The distribution of the entire set of 42 samples was performed by principal components analysis (PCA) and hierarchical clustering (HC). The coal samples were divided into two clusters, numbered I and II. Cluster I completely matched the samples included in subset A (blends), while cluster II matched the samples in subset B (single coals). A basic statistical evaluation of the investigated properties in both clusters I and II was performed, including correlation and regression analyses. Based on the results of FA, the reduced number of 9 relevant characteristics was selected. These were then gradually reduced from 9 to 3; HC separations were calculated for each of them. It was found that almost the same differentiation of 42 samples into clusters I and II (corresponding to blends and single coals, respectively) can be calculated using only 7 instead of the original 16 properties. These properties are TPA, Ad, Vdaf, Rr, Vitr, Lipt and b.