Parametric design of curved hydrocyclone and its optimization based on artificial neural networks prediction

In order to obtain a higher particle separation efficiency, a method of parametric design of the cylinder cone section of curved hydrocyclone based on control points and data points is proposed. The radial diameters were used as input variables to the BP neural network to predict separation performance, which makes it feasible to optimize the profile of the cylinder and cone section of the curved hydrocyclone. The separation efficiency of Bezier-Curved-Cone hydrocyclone (BCCH) designed using control points is up to 69.24 %, and the separation efficiency of Spline-Curved-Cone hydrocyclone (SCCH) designed using data points is up to 73.30 %, both of which are higher than that of the conventional Thew's class hydrocyclone (with a separation efficiency of 62.43 %). Meanwhile, the pressure drop of BCCH is the lowest among all of the five hydrocyclones. No.10 hydrocyclone was predicted to be the best one with separation efficiency up to 78.41 % using BP neural network, which was experimentally verified. The novel hydrocyclones with curved profile provides new approach to enhancing separation performance, and the research costs can be reduced by using neural network-based performance prediction and optimization.