Prediction of Coke CSR from Coal Blend Characteristics using Various Techniques: A Comparative Evaluation

Abstract At Tata Steel, laboratory tests are carried out to see the suitability of the coal blends using imported coals from new sources in coke making and for evaluating the quality of coke produced. The present work is aimed to fulfill the need of a model that will predict the coke properties from coal blend characteristics so that optimization of coal blends for producing desired quality of stamp charged coke could be done easily, quickly, and with a lesser number of carbonization tests in a 7 kg test oven. CSR is predicted with reasonable accuracy from 8 coal blend characteristics (ash, volatile matter, average vitrinite reflectance, crucible swelling number, total reactives, total inerts, vitrinite distribution [V9–V13], and Basicity Index), using different statistical analysis tools (MLR and PCR) and the ANN technique. ANN using the MLP network was found to be most suitable technique for coke properties prediction followed by PCR and MLR. Keywords: Basicity indexBayesian networkCRICSNCSRFluidityInertMLPPCAPCRRBFReactiveRegressionStamp charging Notes Note: Bulk density: 1150 kg/m3; Oven Temperature: 900 ± 5 °C; Moisture: 10%;Carbonization time: 5 hours; Crushing fineness of coal: 90% below 0.0032 m. Ind. Med. Coking – Indian Medium Coking Coal; Imp. 1 – Imported Coal 1; Imp. 2 – Imported Coal 2; Imp. 3 – Imported Coal 3. Test and validation Results: • Actual ▪ Predicted. Training Results: ——— Actual … … Predicted.