Application of artificial neural networks and response surface methodology approaches for the prediction of oil agglomeration process

Oil agglomeration can be a promising technique to recover fines and ultra-fines coal particles from the discarded tailing generated from coal preparation plants. In the present study, an artificial neural network (ANN) and response surface methodology (RSM) was used to predict the behavior of coal oil agglomeration in terms of % ash rejection (% AR) and % combustible matter recovery (% CMR). A three layered Feed Forward Neural Network was developed by varying process variables such as solid concentration (SC), oil dosage (OD) and agglomeration time (AT). Waste soybean oil was used as bridging liquid. An approach of Multilayer Forward Back Propagation Neural Network has been used in conjugation with the hyperbolic tangent sigmoid (tansig) as transfer function. The network is well trained (learning) with Levenberg-Marquardt (LM) algorithm. For further improvement in the generalization of the developed ANN model Bayesian regularization technique has been adopted. Sensitivity analysis was performed using Garson's algorithm, Pearson correlation coefficient and Connection weight approach. The % CMR values predicted from ANN have been in good agreement with the obtained experimental values (R-2 0.9965) and shows better correlation between predicted and observed values than RSM (R-2 0.9892). Also, for % AR the correlation (R-2 0.9965) obtained using ANN found to be higher than RSM (R-2 0.9956).