Research And Implementation Of Performance Prediction Of Centrifugal Pump Based On Neural Network

The research of pump performance prediction is based on the geometric parameters of pump impeller, volute, guide vanes in pumps to analyze the internal flow characteristics so as to predict the performance of pumps. It is an indispensable part of the pump design. This research can help to shorten development cycles, reduce development costs and improve product design qualities. Therefore, pump performance prediction research has been a very important topic for pump experts. However, so far, the research results still have low precision, which can’t meet the actual needs of engineering deficiencies. So researches on the performance of centrifugal pumps not only have important academic value and social benefits, but also provide the basis upon which to draw for other pump performance studies. The main contents and results of this paper are :1.The researcher summarized the status quo, especially various types of centrifugal pump performance prediction of the current domestic and international trends. According to the merits and dismerits of hydraulic loss method, the flow field calculation method and neural network method, the researcher determined the research contents.2.The researcher analyzed the influence of various parameters of centrifugal pumps, which lays the foundation for the research behind.3.By description of Matlab software, the researcher analyzed Matlab software’s advantage on data processing, nonliner fitting, and dynamic simulation. As Matlab neural network toolbox can achieve the nonliner intelligent control, so the researcher took neural network control as research topic.4.Two algorithms based on neural networks: namely Matlab achieved using VC6.0 and Bayesian BP and GA-RBF neural network algorithm to predict the performance of a centrifugal pump designed to achieve.5.The researcher established a GA-RBFneural network model. According to National Excellent Hydraulic Model Compilation published by Shengyang Institute of Pumps, and Modern Pump Technology Manual written by Professor Guan Xingfan, Jiangsu University. The researcher selected 57 groups of design parameters and test parameters of single-stage single-suction centrifugal pumps. By modeling and verification, the researcher found that GA-RBF neural network model is equally effective with RBF neural network model. It is much easier and more convenient to set parameters.6.The researcher established a Bayesian-BP neural network model. According to National Excellent Hydraulic Model Compilation published by Shengyang Institute of Pumps, and Modern Pump Technology Manual written by Professor Guan Xingfan, Jiangsu University. The researcher selected 57 groups of design parameters and test parameters of single-stage single-suction centrifugal pumps. By modeling and verification, the researcher found that BRBP neural network model is equally effective with LMBP neural network model and RBF neural network model, and it is much easier and more convenient to set parameters. It is also a very promising method for centrifugal pump performance prediction.