A Methodology for Automated Pellet size Distribution in a Pellet Mill

In this work the feasibility of pellet size distribution using images captured from the operator’s sight glass of the moving pellet bed using a standard digital camera is investigated. The pellet size distribution is determined using a bespoke circle detection program which is able to fit circles to pellet edges, neural networks are used to match predicted distributions to actual distributions and discrete element modelling is used to create a large quantity of rendered images with known distributions for network training. The results show that the circle detection algorithm is able to find a high percentage of pellets in an image and find their sizes. Neural networks are then able to use the predicted pellet size distributions to give a good prediction of the actual distributions and are able to account for segregation effects in the pellet beds. This demonstrates that an online pellet sizing scheme is feasible in nickel pellet production.