Neural Networks in Dynamic Process Analysis
Abstract
This article presents a study performed within the framework of an industrial research project subsidised by the ECSC. Methods of signal encoding and analysis based on the dynamic evolution of signals coming from an industrial process are presented. They are based on self-organising maps of Kohonen, a well known neural network family for clustering. The main advantage of this approach is that the basic components used for the encoding are based on a learning algorithm, thus the encoding is well adapted to the analysed signals. An other advantage is the extensibility of the presented method to multi-variable analysis. First results are shown.
Keywords
process diagnosis
réseaux de neurones
artificial neural networks
clustering
self-organising maps
pattern recognition
time series
signal analysis
signal encoding
artificiels
classification
cartes auto-organisatrices
reconnaissance de formes
série temporelle
analyse de signaux
codage de signaux
diagnostic de processus