01. Experimental/Computational Fluid Dynamics

The impeller shape of a very low-specific speed pump obtained by applying the design method for general specific speed pumps will increase the friction loss in the flow channel between the vanes. Therefore, radial vanes are commonly used for the low-specific speed pump because the disk friction loss can be reduced by the small diameter impeller which is possible by increasing the angular momentum change. In this case, however, the head-flow rate curve has an upward-sloping curve characteristic in the low flow rate range, which is known to cause unstable conditions. The impeller shape with a downward-sloping head curve characteristic can be estimated from Euler's equation. Therefore, in this study, the impeller shape of the low-specific speed pump having a higher efficiency and downward-sloping head characteristic was sought by the optimum design method. A three-layered neural network (NN) is used as the response surface used in the optimum design method. Pump efficiency, net head, and minimum static pressure around the impeller inlet are calculated by the results of CFD. The required performance is as follows: (1）High efficiency. (2) The net head at the design point is made less than that in the closing operation, which can be calculated from the absolute velocity at the impeller exit assuming the same with the peripheral velocity. (3) The static pressure around the impeller inlet should be not too low to prevent cavitation. We defined an evaluation function that can comprehensively evaluate the above-mentioned three target values. The NN is trained so that the evaluation function value obtained by CFD becomes the output from the output layer, when the shape parameters of the impeller are input to the input layer. The trained NN has the function of a response surface, it was used to search for the combination of shape parameters. Then, the process of optimum design method with CFD was repeated until impeller geometry has a sufficiently high value of the evaluation function.