Abstract: In view of the fact that the performance of Variational Mode Decomposition (VMD) is easily affected by the setting of mode number, penalty coefficient, and the selection of Intrinsic Mode Function (IMF), while the existing studies mostly optimize the parameters based on the properties of the signals, which cannot directly correlate with the strength of the fault characteristics, an inverse parameter optimization method is constructed. First, the optimal mode number is determined based on the intersection of the fast-declining phase and the smooth phase of the center frequency difference curve. Then, a fault characteristic frequency saliency index is proposed. Based on this index, the optimal penalty coefficient and IMF are obtained by inverse optimization. Finally, the optimal parameters are used as the inputs of VMD for fault diagnosis. Applying the method to rolling bearing fault diagnosis, the validity is verified by the rotor-rolling bearing-casing tester and the aircraft accessory gearbox tester. The results show that compared with multiple parameter optimization methods, the proposed method can extract more significant fault features with higher global search capability and computational efficiency.