NEURAL NETS FOR THE SIMULATION OF MINERAL PROCESSING OPERATIONS .2. APPLICATIONS

This paper shows that neural nets exhibit exceptional promise as model ling tool and can be applied and developed further for various applica tions in the metallurgical processing industry. It is described how a sigmoidal backpropagation neural network (SBNN) model for the classifi cation efficiency of a hydrocyclone classifier can be developed on the basis of sufficient data. However, data are expensive and difficult t o obtain for many systems in the processing industry. As difficulties are encountered if a nonparametric model is constructed on the basis o f sparse data, a new neural network modelling technique is described t o obviate this problem. The hybrid subspace method has been developed to isolate the dimensions of less-significant variables and to identif y some mathematical relations, so that the ill-defined dimensionality is reduced and the population density of data is increased accordingly . It has been found that the performance of a hybrid subspace model fo r the kinetics of a typical processing operation is superior to that o f an SBNN model for the entire predictor variable space.