Modelling of oil agglomeration of dolomite by the use of an artificial neural network optimized using Optimal Brain Damage algorithm

This paper investigates application of Artificial Neural Network (ANN) for dolomite oil agglomeration modelling including parameters such as surfactants concentration, oil dosage, time, pH and mixing intensity. The main algorithm implemented for weights calculation was the Levenberg-Marquardt (LM) method. Common problem during process design of neural models is suitable selection of structure complexity. It is known that several connection can influence on final results after off-line training. For improvement of this stage of preparation of the net, pruning method was implemented. Analysed algorithm was based on the main theory of Optimal Brain Damage (OBD) technique. Results present high quality of process recovery prediction. Achieved outcome also shows that reduction of the applied of the neural network can lead to higher precision of calculation. (paper)