NEURAL NETS FOR THE SIMULATION OF MINERAL PROCESSING OPERATIONS .1. THEORETICAL PRINCIPLES

The ill-defined nature of processes in the metallurgical industry nece ssitates the quest for new modelling techniques to emulate features of processes which are poorly understood from a fundamental point of vie w. For this reason nonparametric regression techniques such as neural nets offer an appealing alternative to fundamental modelling. The robu st associative and computational properties of neural networks make th ese regression tools ideally suited for the modelling of ill-defined s ystems. Being the most commonly-used connectionist network, sigmoidal backpropagation neural networks (SBNN's) have been shown to model meta llurgical and chemical systems satisfactorily without any a priori kno wledge about the system provided sufficient data are available. This p aper introduces the field of connectionist networks to the metallurgic al process engineer and describes the fundamentals of an SBNN.