Abstract: To overcome inaccurate results from fault feature extraction on analog circuits produced by noise and the mode mixing phenomenon of empirical mode decomposition (EMD),a new method for fault feature extraction on analog circuits based on lifting singular value decomposition (LSVD) and ensemble empirical mode decomposition (EEMD) is proposed. First,the noise in analog circuit output signals with random noise under different conditions is removed by LSVD,which eliminates noise effects and strengthens the local signal features. The processed signals are then decomposed by EEMD to extract intrinsic mode functions (IMFs). Finally,the energy entropy of each condition is calculated,which can act as the feature sent to the neural network for fault diagnosis. Simulation results of X-axis servo circuit failure diagnosis of the IMU (inertial measurement unit) show that the fault features extracted by the proposed method can effectively diagnose all faults.