In order to effectively diagnose the internal leakage fault, the internal leakage of EHA double-rod hydraulic cylinders with different failure levels was simulated through the EHA test bench. The pressure signals in the hydraulic cylinder chamber were collected under the same control signal input, and the wavelet and EMD decomposition methods were used to decompose the pressure signal in the cavity with the health mode and the failure mode, the first and second high-frequency wavelet coefficients and the first-level IMF function of the pressure signal were obtained, and the feature extraction was completed. The root mean square values after feature extraction were obtained by multiple experiments, the offline diagnosis of leakage faults and fault levels in the EHA hydraulic cylinder was realized, and the diagnosis effects of different diagnosis methods were compared. The result shows that the first-level IMF function of the pressure signal has the best diagnosis effect, and the first-level high-frequency wavelet coefficient has the poorest diagnosis effect.