The production of equivalent transfer functions from trained networks

Summary form only given, as follows. It may be extremely difficult to produce an analytical mathematical description for many engineering processes. Artificial neural networks have been used to learn the transfer functions of such processes, resulting in better optimization and control. One of the major criticisms of this technique has been that the solution produced is a `black box´ model, with no equation provided for the mapping between input and output spaces. The authors have proposed a technique that enables a trained network to be reduced to a set of equations, one for each output, in terms of its inputs