Fault detection and diagnosis using an art-based neural network

The Generalized Adaptive Resonance Theory (GART) network is a neural network model based on the integration of Gaussian ARTMAP and the Generalized Regression Neural Network. It is capable of online learning, and is effective in tackling classification as well as regression tasks, as demonstrated in our previous work. In this paper, we further enhance the capability of the GART network with the Laplacian functions and with new vigilance and match-tracking mechanisms. In addition, a rule extraction procedure is incorporated into its dynamics, and its applicability to fault detection and diagnosis tasks is assessed. IF-THEN rules can be extracted from the weights of the trained GART network after a pruning process. The classification and rule extraction capability of GART are evaluated using one benchmark data set from medical application, and one real data set collected from a power generation plant. These results are then compared with those reported by other methods. The outcomes demonstrate that GART is able to produce high classification rates with quality rules for tackling fault detection and diagnosis problems.