Modelling the tap density of inorganic powders using neural networks

In the present study, the tap relative density of five inorganic powders is modelled using neural networks. These powders are similar in shape but have different true density. A large number of mixings are prepared from three classes (coarse, medium, and fine particles) and modelled. The inputs of the neural networks are the 23 weight percentage intervals of the grain size distribution (38–2000 μm). The estimated values are compared to those obtained by factorial plans. It is shown that very accurate results are obtained with a unique relatively small neural network. Finally, the neural network is used to determine the mixing leading to the highest tap relative density.