Neural and fuzzy model performance evaluation of a dynamic production system

Although mathematical modelling techniques are very well developed, some production processes are difficult to be modelled by these modelling techniques or their math-models are too complex to be used for real-time control due to uncertain, imprecise and vague parameters relations. Spray dryers are complex, dynamic and ill-defined production processes. Their product (powder) must have a controllable size distribution consisting of spherical shapes and free-flowing characteristic of particles, which is required for an ideal pressing operation to overcome the product sticking in the dies. The relations of production process' parameters are highly non-linear. In this study, these non-linear parameters were studied and three different soft-computing intelligent models were developed and used to predict uncertain parameter relations. The first is the fuzzy model of the production process; the others are the artificial neural network (ANN) architectures; the back-propagation multilayer perceptron (BPMLP) algorithm and the radial basis function network (RBF). To deal with uncertainty and vagueness of the production system, a method (methodology) based on a fuzzy hierarchical analytic process modelling approach and two ANN approaches was applied. The performance of the BPMLP algorithm was found most vigorous than the RBF and fuzzy modelling approach.