The multi-stage adding strategy on Kriging and applied to cavitation optimization of centrifugal impeller

In order to realize the intelligent optimization design of cavitation performance of fuel centrifugal pump, a multi-stage active learning adding strategy based on Kriging is proposed. The focus is to establish the three-stage adding strategy by using MSE, EI and CV, and clarify the switching criteria for each stage. Based on the test functions such as one-dimensional function and multi-dimensional function, it is compared with the single-stage adding strategy to verify the effectiveness of the proposed strategy. Furthermore, for a certain type of fuel centrifugal pump, the optimization application of the proposed strategy in its cavitation characteristics is completed. The result of the test function calculation shows that the proposed strategy takes slightly longer to achieve the same accuracy than MSE, which is better than EI and CV. However, compared with MSE, the local accuracy of the global optimal solution is higher. The result of centrifugal pump case shows that the pressure loss at the inlet of the optimized pump decreases, the local flow such as return flow and vortex at the inlet is weakened, the pressure gradient and the vortex flow in the volute weakens. Moreover, the cavitation coefficient of the optimized pump is increased by 30.83%, the NPSH allowance is reduced by 9.14%, the cavitation ratio is reduced by 7.17%, and the gas content in the pump is reduced by 17.27%. Thus, the proposed method improves the cavitation performance of the centrifugal pump effectively.