Optimization design of centrifugal pump impeller based on multi-output Gaussian process regression

Abstract

To overcome the problems of large calculation cost and high dependence on designers’ experience, an optimization design method based on multi-output Gaussian process regression (MOGPR) was proposed. The hydraulic design method of centrifugal pump based on the MOGPR model was constructed under Bayesian framework. Based on the available excellent hydraulic model, the complex relationship between the performance parameters such as head, flow rate and the geometric parameters of centrifugal pump impeller was trained. The hydraulic design of the impeller for M125-100 centrifugal pump was performed by the proposed MOGPR surrogate model design method. The initial MOGPR design was further optimized by using the proposed MOGPR and NSGA-II hybrid model. The initial sample set for NSGA-II was designed by Latin hypercube design based on the MOGPR initial design. The relationship between the impeller geometry and the CFD numerical results of the sample set was trained to construct the surrogate model for pump hydraulic performance prediction. The MOGPR surrogate model was used to evaluate the objective function value of the offspring samples in NSGA-II multi-objective optimization. The comparison of the pump hydraulic performance between the optimized designs and the initial design shows that the efficiency and the head of the tradeoff optimal design are increased by 2.5% and 2.6%, respectively. The efficiency of the optimal head constraint design is increased by 3.2%. The comparison of the inner flow field shows that turbulent kinetic energy decreases significantly and flow separation is effectively suppressed for the optimal head constraint design.