Improving slip factor prediction for centrifugal pumps using artificial neural networks

Slip factor is an important parameter in the performance prediction of centrifugal pumps. This paper presents a new approach improving slip factor estimation of centrifugal pumps. In order to understand the effects of geometrical parameters of impeller (i.e. blade number, blade outlet angle, radius ratio, and blade turning rate) on slip factor, a set of computational fluid dynamics based analyses were carried out for over 70 different impellers by means of a commercial code, CFX. The numerical model was validated with the available experimental data. Subsequently, a multilayer feed-forward neural network, as a popular neural network based fitting tool, was used to generate a mapping between blade geometrical parameters as input variables and slip factor as output variable. The ANN (i.e. artificial neural network) model generalization capability was assessed by comparing the network predictions with those calculated from the test data. The comparative assessment of different slip models showed that the neural network model has succeeded in improving the accuracy of previous slip factor models for centrifugal pumps.