Performance Reliability Analysis of Centrifugal Pump Under Dimensional Deviations Based on Bayesian‐Kriging‐Copula

Abstract

To evaluate the performance distribution of centrifugal pumps under dimensional deviation in design, this paper proposes a novel reliability analysis method combining the Bayesian‐Kriging model and the Copula function. A data‐driven surrogate model is established to rapidly predict the performance responses corresponding to varying design parameters. Both the manufacturing uncertainty of design dimensions and the epistemic uncertainty of the surrogate model are fully considered, and the Bayesian posterior distribution is employed to propagate the epistemic uncertainty of the surrogate model. The Copula function is adopted to construct the joint distribution of two critical pump performance indicators, namely head and efficiency, which effectively reduces the required number of sampling points. The developed Bayesian‐Kriging surrogate model achieves a relative error of less than 1% for all test samples, demonstrating satisfactory prediction accuracy. The joint reliability evaluation results obtained by the Copula‐based method are highly consistent with those calculated via the Monte Carlo sampling (MCS) method. For a single test case, the MCS method requires 10 8 samples, whereas the proposed Copula‐based reliability method only needs 10 3 samplings, which significantly reduces the computational cost. These results verify the outstanding advantages of the proposed method in terms of both calculation accuracy and computational efficiency. Furthermore, performance reliability evaluation of a typical centrifugal pump is conducted using the developed method, yielding a reliability value of 0.99820 with a 95% confidence interval of [0.99806, 0.99833].