Fault Diagnosis Method of Synchronous Condenser Based on Data-Driven and Multisource Data Fusion

Synchronous condensers provide strong short-term overload capacity and instantaneous reactive power support, playing a crucial role in addressing issues such as insufficient reactive reserves, reduced short-circuit capacity, and commutation failures in ultra-high-voltage direct current (UHVDC) transmission systems. However, rotor inter-turn short circuit (ITSC) faults can lead to severe consequences, including increased vibrations, magnetic field distortion, and even forced shutdown. To improve the accuracy and robustness of fault detection, this paper proposes a data-driven fault diagnosis method based on multi-source data fusion. Excitation current, radial leakage flux, and electromagnetic noise are selected as non-invasive diagnostic signals. To mitigate the impact of noise and non-stationary characteristics in the acquired signals, a Correlation Variational Mode Decomposition (CVMD) algorithm is employed to denoise and extract effective signal components prior to feature processing. The decomposed signals are then fed into a Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) model for deep feature learning and classification. Experimental validation using an MJF-30-6 salient pole synchronous motor demonstrates that the proposed method significantly enhances diagnostic performance, achieving a fault identification accuracy of 99.3%, and effectively overcomes the limitations of single-signal diagnosis under complex operating conditions.