In order to study the second-order simulation method under non-design operating conditions and explore the feasibility of real-time simulation interaction of digital twins,a 14-stage centrifugal pump was used as the research object,the Latin hypercube sampling method was adopted to select sample operating points randomly,and a multi-stage centrifugal pump digital twin steady-state ROM was trained that satisfied the root mean square(RMS)error accuracy condition of less than 1％.The consistency between the ROM and the FOM numerical simulation results was tested by comparing the distribution of internal flow field pressure,velocity,and vortex structure.The research results demon-strate that the multi-stage pump ROM trained based on the proper orthogonal decomposition method can accurately and quickly predict and calculate the head and internal flow field pressure distribution under specific operating points.The average relative error near the impeller is less than 2％,which is suffi-cient to meet the needs of fast simulation in practical applications.The velocity distribution at the im-peller rotor-stator interface of ROM is consistent with that of the FOM,but the sensitivity to velocity changes in the area with large velocity gradients is insufficient.As the number of training samples in-creases,the accuracy gradually improves.ROM can accurately estimate the shape of large-scale vortices in the impeller,but due to the omission of high-order information during the model reduction process,the vortex edge obtained by ROM is smoother and smaller in volume than the FOM calculation results,with an average relative error of about-7.8％.This study verifies the effectiveness of the POD method ROM in the internal flow field of a centrifugal pump and establishes a computational foundation for the construction and real-time simulation of digital twins in the field of fluid machinery.
关键词
多级离心泵/数字孪生/降阶模型/本征正交分解/流场分析
Key words
multi-stage centrifugal pump/digital twins/reduced-order model/proper orthogonal decomposition/flow field analysis
维普
万方数据