Abstract
This paper proposes a fault diagnostic method for pump running conditions. The method is a combination of a process modeling and a classification procedure. The pump head and hydraulic losses in the pipe system are modeled by two equations. The values of the coefficients in the equations are determined from measurable output variables of the pump. Since the pump running conditions affect the coefficient values, they are detected by classifying those values. A multi-layer neural network is employed for the classification. Three running conditions of a drainage pump are clearly detected by this method. The accuracy of detection is improved by increasing the hidden layers and their units in the neural network.