Prediction research on cavitation performance for centrifugal pumps

Abstract

The present situation about cavitation performance prediction of centrifugal pump is introduced. The primary methods of cavitation performance prediction for centrifugal pumps are summarized, including numerical simulation method and artificial neural network method. Based on the neutral network toolbox of MATLAB7.0, topological structures of artificial neural networks are determined and network models for predicting cavitation performance of centrifugal pumps are established by analyzing the relations between geometric parameters of centrifugal pumps and net positive suction head at designed flow rate, The BP and RBF neural networks are trained by 60 example data, which are obtained from engineering practice and normalized by using neural network toolbox function. The cavitation flow in centrifugal pumps is simulated by using the commercial CFD code FLUENT6.2 A moving reference frame technique is applied to take into account the impeller-volute interaction. The standard k-? turbulence model, mixture multiphase model and SIMPLEC algorithm are used. Velocity inlet and pressure-outlet are set as boundary conditions.The cavitation performance curves at design condition are predicted by calculating the head under different net positive suction head. The cavitation performances of 3 pumps with the different specific speeds are predicted by using numerical simulation method and neural network method respectively. The predicted values are compared with the tested values, the results show that the predictions by two methods are satisfied, the relative declination of BP and RBF for 3 pumps are 2.87?, 2.55?, 5? and 3.71?, 3.27?, 4.62? respectively. The absolute declinations of numerical simulation method are 0.17m, 0.08m and 0.16m. The advantage and disadvantage of those two methods are compared, The numerical simulation method will take a lot of time to modeling and calculating, but the law of cavitation flow in the centrifugal pumps can be obtained, which are helpful to disclosing the mechanism of cavitation characteristic; The artificial neural networks method needs a great deal of training examples, which are necessary and important to the prediction accuracy, but the math relation between input variables and output variables can be set up by using artificial neural network method, which is useful to optimize the structure of pumps.