Multi-component optimization of a vertical inline pump based on multi-objective pso and artificial neural network

Abstract

The vertical inline pump is a single-stage single-suction centrifugal pump with a curved inlet pipe before the impeller, which is widely used in where the constraint is installation space. In this paper, with the objective functions of efficiencies at 0.5Qd, 1.0Qd, and 1.5Qd, a multi-objective optimization on inlet pipe and impeller was proposed to broaden the efficient operating period of a vertical inline pump. Two 5th order Bezier curves were adopted to fit the shape of the mid curve of the inlet pipe and the trend of the blade angle of the impeller. Fourteen design variables were selected after the data-mining process. 300 sample cases were generated using Latin hypercube sampling (LHS), and they were solved by 3D RANS code to obtain the objective functions. The feed-forward artificial neural network with a hidden layer and an output layer was adopted to fit the two objective functions and the 14 design variables. The Pareto frontiers were generated for the three objectives using multi-objective particle swarm optimization (MOPSO), and three different configurations on the Pareto front are selected for detailed study by computational fluid dynamics (CFD). The results showed that the profile of the inlet pipe and the blade have a dramatic impact on the performance of the vertical inline pump. The Pareto frontiers reported that the performance under the overload condition usually keeps stable when the nominal efficiency is lower than 82 %, or the part-load efficiency is lower than 62 %, and it will decrease rapidly after that. After optimization, the improvement of efficiencies at the part-load condition and nominal condition of the picked case were 9.65 % and 7.95 %, respectively.