The electromagnetic vibration characteristics of doubly fed induction machines (DFIMs) employed in variable-speed pumped storage units, which must accommodate frequent power response and operational mode transitions, serve as critical indicators for assessing unit safety and stability. Nevertheless, there persists a significant research gap regarding generalized vibration analysis models and comprehensive investigations into their steady-state and dynamic vibration performance. To address this challenge, this study develops a universal analytical model for electromagnetic excitation forces in DFIMs using Maxwell’s stress tensor method, explicitly incorporating operational conditions such as rotor eccentricity and load imbalance. Using a 300 MW DFIM as a case study, we employ a hybrid numerical-analytical approach to examine the detrimental effects of harmonic currents generated by rotor-side converters. Furthermore, we systematically analyze how spatial harmonics induced by mechanical faults and temporal harmonics arising from electrical faults collectively influence the electromagnetic vibration behavior. Experimental validation conducted on a 10 MW DFIM prototype through vibration displacement measurements confirms the efficacy of the proposed analytical framework.

Experimental approaches for the identification of dynamic parameters in synchronous machines mainly include two methods, a three-phase sudden short-circuit (TPSSC) test and a standstill frequency response (SSFR) test. However, the former has significant safety risks, while the latter has a complex implementation process, resulting in insufficient adaptability to large-scale units. To overcome the above obstacles, this paper proposes an improved DC-decay test method that can be performed at an arbitrary rotor position so that the rotor pre-positioning process in the conventional DC-decay test can be neglected. Meanwhile, combining the transient analysis theory and particle swarm optimization algorithm, a semi-analytical parameter identification method is proposed. Finally, the proposed method is applied using a 172 MVA large synchronous machine. Compared to the results obtained by the TPSSC test using the Prony algorithm and other conventional type tests, the error of the parameter calculation results obtained with the conventional method reached a maximum of 16.6%, while that of the proposed method was merely 8.6%, and the experimental period could be shortened from 5 days to half a day.