Research On Optimization Of Centrifugal Pump Impeller Based On Dynamic RBF Surrogate Model And Improved NSGA-Ⅱ

Centrifugal pump has important applications in petroleum,chemical industry,aerospace,ships and other fields,but its hydraulic optimization is difficult due to its internal complex flow constraints.In addition,due to the complex relationship of blade shape,hydraulic performance and internal flow structure of centrifugal pump,it takes a long time to obtain the optimization target value,so the surrogate model is usually used to solve the target value.RBF neural network can construct the nonlinear mapping relationship between input data and output data of any problem according to the training sample data,which is widely used in the centrifugal pump optimization.In the process of multi-objective optimization,there are several disadvantages in the traditional NSGA-Ⅱ algorithm,such as the length of iterative time and the large amount of computation.And the traditional RBF neural network surrogate model has the problem that the prediction accuracy will gradually decrease as the predicted samples gradually move away from the initial sample set.In this study,the modified NSGA-Ⅱalgorithm and the dynamic RBF neural network surrogate model strategy was proposed.The modified NSGA-Ⅱ algorithm and the modified RBF neural network surrogate model were used to complete the multi-objective optimization of centrifugal pump.The main research contents are as follows:(1)A strategy of dynamic RBF neural network was proposed,and the modified strategy was verified by test function.The dynamic RBF surrogate model can effectively improve the optimization effect.the MH48-12.5 low specific speed centrifugal pump was regarded as the research object,and 40 samples were designed by Latin hypercube design.The blade shape was generated by the parametric method,and the sample data were obtained by CFD numerical simulation.The static RBF surrogate model was trained to perform the impeller optimization.The multi-objective optimization results of the traditional NSGA-Ⅱ algorithm combining dynamic and static RBF neural network surrogate model were compared and analyzed.The results show that the Pareto front of the dynamic RBF neural network surrogate model is better than that of the static surrogate model.The CFD numerical simulation was used to verify the reliability of the dynamic RBF neural network surrogate model by the maximum head design and the maximum efficiency design.The results show that the dynamic RBF surrogate model is better than the traditional static RBF surrogate model.(2)The NSGA-Ⅱ algorithm based on parent partition strategy was proposed.The multi-objective effect of the modified NSGA-Ⅱ algorithm was verified by using the test function.The modified NSGA-Ⅱ algorithm can obviously accelerate the iteration speed and reduce the calculation time.The MH48-12.5 low specific speed centrifugal pump impeller was optimized by the multi-objective method.The Pareto front of the modified NSGA-Ⅱalgorithm is basically coincident with the static surrogate model,but the modified NSGA-Ⅱalgorithm reduces the number of iterations and saves time.The CFD numerical simulation was used to verify the reliability of the modified algorithm by the maximum head design and the maximum efficiency design.(3)The multi-objective optimization of the MH48-12.5 low specific speed centrifugal pump was carried out by using the NSGA-Ⅱ algorithm which is based on parent partition strategy and dynamic RBF neural network strategy,whose results show that it can be used for multi-objective optimization of centrifugal pump.Compared with the traditional NSGA-Ⅱ and dynamic RBF neural network strategy,its Pareto front is basically consistent,the number of iterations is obviously reduced,and the optimization time is saved.Compared with the traditional NSGA-Ⅱ algorithm and the RBF surrogate model,it can effectively improve the optimization effect,reduce the time and computation of multi-objective optimization design,and verify the maximum head design and the maximum efficiency design by CFD numerical simulation.The results show that the head of the optimal head design obtained is 3.60%higher than that of the prototype pump and 1.35% higher than head design obtained by original algorithm.The efficiency of the optimal design obtained by the modified algorithm is4.90% higher than that of the prototype pump and 1.82% higher than that of the optimal design original algorithm.The reliability and superiority of the NSGA-Ⅱ algorithm which is based on parent partition strategy and dynamic RBF neural network strategy were verified.(4)The NSGA-Ⅱ algorithm based on parent partition strategy and dynamic RBF neural network strategy was selected to obtain the maximum head design and the maximum efficiency design of Pareto front for CFD numerical simulation calculation.And compared with the external characteristic curve,turbulent kinetic energy,pressure,speed and matching degree of the original design and the optimized design,the results show that the optimized design is better than the original design.