A multi-layer feed-forward neural network was trained (using the error
back-propagation algorithm) to learn a subset of rules designed to
stabilise the performances of the copper flotation section of a complex
sulphide processing plant and then tested in order to evaluate its expert
performances. Two types of neural networks were compared: the classical
ones, fully connected between one layer and the adjacents (with different
architectures), and entropy networks, partially connected and generated
according to a methodology that simplifies its building up. The
advantages and drawbacks of both were analysed and compared with
classic (rule based) expert systems.