Optimization Design of the Elbow Inlet Channel of a Pipeline Pump Based on the SCSO-BP Neural Network

Abstract

A vertical pipeline pump is a type of single-stage, single-suction centrifugal pump with a curved elbow input. The inhomogeneous flow of the impeller inlet coexists with the unique elbow inlet channel, making it simple to generate the inlet vortical secondary flow. This paper aimed to optimize elbow inlet channel performance using a backpropagation (BP) neural network enhanced by the Sand Cat Swarm algorithm. The elbow flow channel’s midline and cross section shapes were fitted with a spline curve, and the parametric model of the curve was then constructed. Nine initial variables were filtered down to four optimization variables using the partial factor two-level (P2) and Plackett-Burman (P-B) experimental designs and multivariate analysis of variance. The sample space was generated by 50 groups of experiment samples, and the Sand Cat Swarm algorithm to optimize the BP (SCSO-BP) neural network and the approximation model of four variables were built. A genetic algorithm (GA) was applied to determine the optimal parameters among the approximate models in the sample space, and the ideal parameter combination of the elbow inlet channel was achieved. The findings demonstrated a strong agreement between the experimental and numerical simulation results. With reduced error fluctuation in inaccuracy and a more consistent fluctuation range, the approximate prediction model based on the optimized Sand Cat Swarm algorithm performed better. The optimized inlet model minimized the impact loss on the inlet wall, improved the velocity distribution uniformity of the inlet impeller, increased the pump efficiency by about 5% and the head by about 7.48% near the design flow, and broadened the efficient region of the pump.