In light of the problem to distinguish multiple engine faults through the same method using a single channel signal, the existing algorithms are optimized to extract fault characteristics from vibration signals. First, in view of the difficulty in selecting the decomposition level of the variational mode decomposition (VMD) decomposition levels selection, the initial value of the center frequency iteration is optimized based on the frequency characteristics of several different types of faults, which improves calculation efficiency and convenience while ensuring accuracy. Then, the robust independent component analysis (Robust ICA) is introduced to analyze different signal sources in the same frequency. The fourth-order cumulant of the restructured signals from VMD and Robust ICA is taken as failure indexes. Finally, the cluster center determined by fuzzy C-means clustering is used as the reference point. The Euclidean distance between each test points and the center is used to distinguish fault types. The results show that this method achieves high recognition rate.