Soft Computing Tools in Anytime Systems

Nowadays there could be found more and more applications, where the computations must be carried out "on-line" with a guaranteed response time good examples are the modern diagnostic and monitoring systems. Moreover, the available time and resources could be not only limited, but could also change during the operation of the system. Consequently, so called "anytime" systems are needed, which are able to provide short response time, and maintain the information processing even in case of temporary shortage of time or computational power. On the other hand, fuzzy or neural models could be well used for modeling and fault diagnosis, and they are usable even, when the exact mathematical description of the industrial process is not known, but several measured data and/or symbolic expert knowledge is available. Unfortunately fuzzy and neural models often has a high complexity, because the necessary number of antecedent fuzzy sets or neurons for the desirable accuracy could not be exactly determined. In the following, a possible method will be shown for the application of fuzzy inference systems and special neural networks in anytime systems with a modular architecture, with the help of the Singular Value Based Decomposition (SVD).