H. Ben´tez-P´rez, J. Solano-Gonz´lez, F. Cardenas-Flores, and D.F. Garc´a-Nocetti ı e a ´ ı
Fault diagnosis, time-varying systems, ART2A networks
Fault diagnosis currently offers different alternatives to classify faults at early stages, such as model-based and knowledge-based techniques. Nevertheless, fault classification for time-varying systems is still an open problem. Strategies such as self-organizing maps and principal component analysis ensure fault classification to bounded time-variance faults. The approach presented in this paper proposes the use of three non-supervised neural networks. The first two networks overlapped by certain time shift. The third network performs a comparison between the two networks outputs in the previous stage. As a result, the system classifies the fault even if the system is time-variant. The strategy named as Overlapped ART2A Network, aims to obtain an autonomous performance and on-line fault classification. Results show the effectiveness of the approach considering a case study with fault and fault-free scenarios.
Important Links:
Go Back
