Diagnosis method of centrifugal pumps by rough sets and partially-linearized neural network

When using a neural network (NN) for automatic diagnosis, it is difficult to deal with the ambiguous diagnosis problems. The paper proposes the “partially-linearized neural network (PNN)” by which failure types can be quickly distinguished on the basis of the probability distributions of symptom parameters. The knowledge acquisition method for PNN learning using rough sets is also proposed. The authors have applied these methods to the failure diagnosis of centrifugal pumps which are used in a chemical plant. The results of the failure diagnosis have verified that the methods are effective. The methods discussed can also be applied to other diagnosis or pattern recognition problems