Addressing the challenge of extracting fault characteristics from vibration signals of inter-shaft bearings in aeroengine amidst complex transmission paths and strong background noise, this paper proposes a method based on local energy density (LED) for fault feature extraction and diagnosis. Initially, singular spectrum analysis is employed for preliminary noise reduction of the fault signals and optimal reconstruction order determination using a cosine-based approach to preserve crucial fault information within the signal. Subsequently, a novel metric, LED, is introduced to quantify the energy ratio of fault characteristic frequencies and their harmonics within a local frequency range. This metric not only effectively extracts subtle fault features but also demonstrates robustness against deviations between actual and theoretical fault frequencies. Utilizing LED as the fitness function, the method enhances fault features in the denoised signal through maximum correlation kurtosis deconvolution (MCKD) optimized by the artificial hummingbird algorithm. Fault diagnosis is achieved through envelope spectrum analysis. The effectiveness of the proposed method is validated through intermediary bearing fault simulation and noise addition experiments, showing a 20. 7% to 218% increase in the fault feature coefficient (FFC) and a 22. 9% to 134% increase in LED compared to existing fault diagnosis techniques. The method accurately identifies the characteristic frequencies and harmonics of outer race faults under noise conditions of 0 dB, -4 dB, and -10 dB, indicating that the proposed SSA_MCKD can effectively reduce the influence of signal noise and extract fault features of rolling bearing.