Neurofuzzy network

Fig. 9 Schematic representation of neurofuzzy network within ASMOD algorithms. (From Ref.ti l)...

Using the information about the submodel output, and the number of membership functions used for each input, rules can be extracted from the model. A typical example is shown in Fig. 10, where the submodel output shows a non-linear form that is fitted by three membership functions LOW, MID, and HIGH. The rules are shown in the figure the numbers in brackets represent confidence levels and are related to the weights determined by the neurofuzzy network. 

Keywords distillation control, neurofuzzy networks, soft sensors, genetic algorithms... 

Neurofuzzy eontrol eombines the mapping and learning ability of an artifieial neural network with the linguistie and fuzzy inferenee advantages of fuzzy logie. Thus... 

In a series of papers, personnel from Novartis and the University of Basel in Switzerland have highlighted the pros and cons of neural networks for immediate release tablets [37-40]. In other studies neural networks have been found useful in modeling tablet formulations of antacids [41], plant extracts [42], theophylline [43], and diltiazem [44]. In a recent paper Lindberg and Colbourn [45] have used neural networks, genetic algorithms, and neurofuzzy to successfully analyze historical data from three different immediate-release tablet formulations. 

Pigmented film coating formulations have recently been modeled and optimized to enhance opacity and reduce film cracking with neural networks combined with genetic algorithms [46, 47] as well as being studied with neurofuzzy [48]. In the latter study the rules discovered were consistent with known theory. 

Essentially, the neurofuzzy architecture is a neural network with two additional layers for fuzzification and defuzzification. The fuzzification and input weighting are illustrated graphically in Fig. 9, adapted from the thesis of Bossley. It can be seen that there are similarities with the RBF network, except now the radial functions are replaced by the multivariate membership functions. 

Gunn, S. R., Brown, M. and Bossley, K. M. (1997). Network Performance Assessment for Neurofuzzy Data Modelling, Lecture notes in computer science, 1280, pp. 313-323. 

---