A neural network approach for centrifugal impeller inverse design

In this paper a neural network approach is proposed to solve an inverse design problem of a centrifugal impeller when the basic structure parameter and the hub-shroud contours are known, and the expected blade surface velocity distribution is given. The proposed neural networks have a four-layered feedforward architecture and are trained with finite samples by means of a back-propagation algorithm. The simulations show that the trained networks can yield a blade shape that generates the expected velocity distribution on its surface.