Experimental Studies on the Optimization Design of a Low Specific Speed Centrifugal Pump

Abstract

For a low specific speed centrifugal pump with the requirement of high efficiency of 68% and non-overload power characteristics, series experimental studies, by matching 9 volutes with 19 impellers were done. By combining the former research results about the splitters and the non-overload theory in centrifugal pump, the theoretical conditions to achieve the property of non-overload in a centrifugal pump with splitters was analyzed, and formulas to estimate the maximum shaft power and its position are derived. Based on the requirement of high efficiency and non-overload, blade outlet angle β2, blade outlet width b2, volute throat are Ft and the inlet diameter of splitters Di were chosen with three levels to design a normal L9 (34) orthogonal test scheme. After the optimized design scheme was determined, and corresponding test was done also, it demonstrates that the experiment purpose was achieved and the design method to combine the splitters and non-overload theory is reasonable, which can get the property of high efficiency and non-overload. A BP artificial neural network (BPANN) model was built to predict the efficiency and head of centrifugal pumps with splitters in MATLAB toolbox. Eighty five groups of test results were used to train and test the network model, where the Levenberg–Marquardt algorithm was adopted to train the neural network model. Five parameters Q, Z, β2, Di, b2 were chosen as the input layer parameters, η and H were the output factors. Through the analysis of prediction results, the conclusion was got that, the accuracy of the BP ANN is good enough for performance prediction. And the BP ANN can be used for assisting design of centrifugal pumps with splitters, which can shorten research time and cost.