Abstract To detect and isolate faults by classification approach the residual-based pattern recognition was investigated because this approach divides feature vectors into two basic categories: one normal mode (normal class) and one or several fault modes (fault classes). In fault diagnosis system designers or engineers should analyse systems in order to define their input patterns, features, classes required for pattern recognition, and corresponding residuals, fault symptoms, faults needed for fault diagnosis, respectively. As a result of this analysis, there will be a set of priori known faults for off-line teaching using a training and a testing data set obtained from a system. In fault diagnosis systems it is impossible to define all faults and to diagnose all would-be faults. To detect a priori unknown faults on-line training of the neural network is applied using unsupervised learning method.
[1]
Xiao Li,et al.
Integrated diagnosis using information-gain-weighted radial basis function neural networks
,
1996
.
[2]
Heikki N. Koivo,et al.
Application of artificial neural networks in process fault diagnosis
,
1991,
Autom..
[3]
Nicolaos B. Karayiannis,et al.
Growing radial basis neural networks: merging supervised and unsupervised learning with network growth techniques
,
1997,
IEEE Trans. Neural Networks.
[4]
Marios M. Polycarpou,et al.
Automated fault detection and accommodation: a learning systems approach
,
1995,
IEEE Trans. Syst. Man Cybern..