Publication detail
The use of fuzzy-neural networks for prediction of the residual lifetime of insulating materials in electric rotary machines.
HAMMER, M.
Czech title
Použití Fuzzy Neuronové Sítě pro Predikci Zbytkové Životnosti Izolačního Materiálu v Elektrických Rotačních Strojích.
English title
The use of fuzzy-neural networks for prediction of the residual lifetime of insulating materials in electric rotary machines.
Type
conference paper
Language
en
Original abstract
When researching electro-insulating materials, searching for new methods of diagnostics and prediction of the lifetime of those insulating materials assumes more and more importance. We are concerned with degradation process of insulating materials used for insulation systems of electric rotary machines windings by artificial intelligence. Our research is focused to residual lifetime prediction of this insulation system. Most usable methods of artificial intelligence are neural networks, fuzzy systems and genetic algorithms. In this contribution we are especially focused to use hybrid systems like fuzzy-neural networks. We have two main types of these networks. First is less difficult and there is used classical neural network with some no-fundamental changes. We only put on two blocks: block of fuzzyfication into the input and block of defuzzyfication into the output. There are some changes in number of neurons in input and output layers. The result is classical neural network with some changes, which works with fuzzy numbers. In the second type of fuzzy-neural network, there are particular neurons represented by fuzzy sets. There are many differences between this network and classical neural network. All the layers are converted for using as a completely fuzzy model. Modules of fuzzyfication and defuzzyfication are included too. We used back-propagation learning algorithm but converted for our need.
Czech abstract
V dnešní době je kladen čím dál větší důraz na včasnou diagnostiku provozního stavu výrobních strojů a zařízení. Na našem ústavu jsme se soustředili na problematiku degradačních procesů v izolačních soustavách elektromotorů a jejich řešení pomocí metod umělé inteligence. Konkrétně je náš výzkum zaměřen na predikci životnosti izolačních materiálů vinutí elektrických strojů pomocí metod umělé inteligence. Mezi nejpoužívanější metody umělé inteligence patří fuzzy systémy a neuronové sítě. Tato práce je zaměřena na řešení problematiky predikce pomocí fuzzy-neuronových sítí, jenž v sobě spojují výhody fuzzy systémů a neuronových sítí. Existují dva základní typy fuzzy-neuronových sítí. První méně složitý typ spočívá v úpravě fungující klasické neuronové sítě připojením modulů fuzzifikace a defuzzifikace, díky kterým tato síť potom pracuje s fuzzy čísly. Vnitřní struktura této sítě ale zůstává nezměněna. Druhý mnohem náročnější způsob tvorby fuzzy-neuronových sítí spočívá v kompletním návrhu jednotlivých vrstev sítě, přičemž každý neuron je tvořen fuzzy množinou. Síť sama ve své druhé vrstvě provádí fuzzifikaci a na výstupu defuzzifikaci. Pro učení sítě bude využit Backpropagation algoritmus ovšem upravený pro fuzzy čísla.
English abstract
When researching electro-insulating materials, searching for new methods of diagnostics and prediction of the lifetime of those insulating materials assumes more and more importance. We are concerned with degradation process of insulating materials used for insulation systems of electric rotary machines windings by artificial intelligence. Our research is focused to residual lifetime prediction of this insulation system. Most usable methods of artificial intelligence are neural networks, fuzzy systems and genetic algorithms. In this contribution we are especially focused to use hybrid systems like fuzzy-neural networks. We have two main types of these networks. First is less difficult and there is used classical neural network with some no-fundamental changes. We only put on two blocks: block of fuzzyfication into the input and block of defuzzyfication into the output. There are some changes in number of neurons in input and output layers. The result is classical neural network with some changes, which works with fuzzy numbers. In the second type of fuzzy-neural network, there are particular neurons represented by fuzzy sets. There are many differences between this network and classical neural network. All the layers are converted for using as a completely fuzzy model. Modules of fuzzyfication and defuzzyfication are included too. We used back-propagation learning algorithm but converted for our need.
Keywords in Czech
fuzzy neuronové sítě, životnost
Keywords in English
fuzzy-neural networks,lifetime
Released
01.07.2004
Publisher
Petrosani (Romania)
Location
Petrosani
ISBN
973-718-026-7
Book
3rd International Conference, ICPE-CA 2004
Pages from–to
121–125
Pages count
4
BIBTEX
@inproceedings{BUT17304,
author="Miloš {Hammer},
title="The use of fuzzy-neural networks for prediction of the residual lifetime of insulating materials in electric rotary machines.",
booktitle="3rd International Conference, ICPE-CA 2004",
year="2004",
month="July",
pages="121--125",
publisher="Petrosani (Romania)",
address="Petrosani",
isbn="973-718-026-7"
}