Soft-computing methods for diagnosis and design of electrical devices

The paper offers a modern approach to the problems of diagnosis and design arising in electrical engineering. The methods are genetic algorithms, fuzzy logic and neural networks. Two kinds of tasks are considered. First, the algorithm which enables creation of an expert system for diagnosis of technical condition of oil transformers is presented. The system is based on real results of chromatography of gases dissolved in transformer oil (DGA) and creates a set of diagnostic rules using fuzzy logic and a genetic algorithm. An alternative method based on a fuzzy controller is also mentioned. In the second part, attention is given to neural simulation of pulse magnetiser for magnetising permanent magnets. Moreover, the problem of identification of the induction motor parameters from available performance data using the equivalent circuit model and soft-computing methods is compactly presented.