The fast neural network (FNN) is an additional modelling algorithm and offers the possibility of modelling complex and locally varied ranges of measurement data with high quality.
Mathematical models are built using a technique which combines the flexibility of neural networks with the interpretability of fuzzy systems. Local neuro-fuzzy models are generated in different ranges and they are combined to a global, in general nonlinear, model valid over the whole operating region. The procedure is based on the Radial Basis Function Network (RBF).
The FNN modelling algorithm can either be used for the target function or for the restrictions or for the measurement data only. During the modelling process the algorithm divides the modelling range into small sections, builds models for this areas before it starts to calculate an overall model. The size of these sections can be defined by the user just as he can determine the local modelling algorithm order.
The advantage of the FNN algorithm becomes effective, when conventional algorithms as polynomials of second or higher order will not produce satisfactory model quality.
The example given here, shows the CAMEO option "Neural Networks" in operation. Here the entire variation range was divided into many small sections, before the modelling procedure calculated a final overall model.