Because the early weak fault diagnosis effect of Feature Mode Decomposition (FMD) is easily affected by the filter length L and the number of mode decomposition n, the Harris Hawk Optimization algorithm is used to optimize the preset parameters of FMD. Firstly, the method takes information entropy as the objective function. Harris Hawk algorithm was used to compare the objective function values of each component signal decomposed by FMD under different preset parameters, and the L and n corresponding to the minimum value were selected as the preset parameters of FMD. Secondly, each Intrinsic Mode Function (IMF) component is reconstructed based on kurtosis criterion. Finally, the FMD-based signal processing workflow enables bearing fault classification through systematic examination of demodulated vibration spectra. The method is verified by the publicly available fault bearing data from the Case Western Reserve University. It shows good anti-noise ability and effective early weak fault diagnosis capability.