The rotor imbalance vibration signal is proved to be nonstationary and carries intense ambient noise. In view of these characteristics, a fault feature extraction method based on fast independent component analysis (FastICA) and improved Hilbert-Huang transform (HHT) was proposed. FastICA method was used to eliminate ambient noise and other factors of the signals, then the noise reduction signal was decomposed by complete ensemble empirical mode decomposition withe adaptive noise (CEEMDAN) and a set of intrinsic mode function (IMF) were obtained. The illusive component selection technique based on correlation analysis was used to decide which the illusive component was and the accuracy of fault information extraction was guaranteed. The simulation experiment results show the effectiveness of the proposed method. The actual test shows that the proposed method can be used to extract the information of rotor imbalance fault signal effectively and accurately. It provides an effective method for this problem.