Two approaches for learning diagnostic knowledge from past process data or simulation data using neural networks and genetic algorithms are described in this paper. In the first approach multi-layer feed forward neural networks are used to extract the relations between observed abnormalities and the corresponding faults. Through training, neural networks can acquire and store diagnostic knowledge as network weights. The trained networks can be used to diagnose faults in that they can associate the observed abnormalities with the corresponding faults. In the second approach, genetic algorithms are used to train diagnostic rules from process data or simulation data. Genetic algorithm based learning starts with a group of initial rules and produces new rules through reproduction, crossover and mutation. More fitted rules are preserved and less fitted ones are abandoned. Through this evolution-like procedure, effective and concise diagnostic rules can be discovered. The proposed approaches can ease the knowledge acquisition task in developing knowledge based diagnosis system. The proposed approaches have successfully been applied to a pilot scale mixing process.
Fault diagnosis, neural networks, genetic algorithms, learning, knowledge based systems.
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).