A neural network-based approach for steady-state modelling and simulation of continuous balling process

Efficiency of plant operations rely heavily on the stable availability of green pellets of desired size and quality. However, agglomeration plant often operate under capacity because of the sensitivity of balling circuits towards even the small perturbation in operating conditions. Though many researchers came up with various models to estimate the behaviour of continuous agglomeration system, there is still scope to develop improved modelling and simulation techniques. In this study, we present a neural network-based approach to simulate the nature of continuous balling process for better circuit control and improved plant efficiency. Mathematical expressions are developed to capture the response of produced and recycled load for a given set of parameters. Using these expressions, a multilayer perceptron model is trained that can predict the behaviour of circuit for pre-specified values of operating conditions. After simulation, effect of varying parameters on the dynamics of produced and recycled mass is summarized. Moreover, variations in process properties such as average recycled load, cycles needed to achieve steady state and maximum amplitude of recycled mass are also discussed.