Multi-Objective Optimization Design Of Centrifugal Impeller Based On BP Artificial Neural Network

As an important role of energy consumption in transportation and chemical industry, centrifugal compressors own great energy saving potential. However, the traditional "trail-error" design methods can not satisfy the demands for higher energy efficiency. Designers need to modify the parameters for many times to meet the design requests, while they may not be able to get the optimal array. BP(back propagation) ANN(artificial neural network) has good applicability, self organization and generalization ability. GA(genetic algorithm) has the advantages of global optimization and the ability to search the optimal solution without gradient information. Both of them are applied to centrifugal compressor impeller design and optimization in this passage. This passage can be divided into two parts:In the first part, the database of the centrifugal compressor impeller performance is given by NREC. The relationship of the impeller main design parameters and performance is established by BP artificial neural system. Then the optimal array of main design parameters is acquired by the multiobjective genetic algorithm. Through the discussion of artificial neural system model’s reliability, multi-objective optimization model’s principal component analysis and the impeller’s 3D numerical simulation verification, the effectiveness of the present centrifugal impeller design method is well confirmed. The optimized impeller’s efficiency and total pressure ratio has increased by 1.4% and 10.9%. The results shows that the ANN model used in this passage fit the data well, it is precise enough to replace the role of CFD in preliminary design process; the Pareto optimal solutions acquired through multi-objective optimization model have preserved the features of the database; GA can find the best solutions for certain demands, based on suitable objective functions.In the second part, Design3D is used to operate the impeller’s parameterization and optimization. The impeller acquired in the first part is improved by this process, both in isentropic efficiency and total pressure ratio. The CFD results shows that the present design approach can provide accurate references for centrifugal compressor designers, the optimized impeller’s efficiency and pressure ratio has increased by 3.7% and 13.6%. And, from some aspect, it revealed the instructive meaning of analyzing statistical laws from experimental data.