In order to effectively diagnose rotating machinery faults, a fault diagnosis and recognition method for multidimensional features extraction based on ensemble empirical mode decomposition (EEMD) was proposed. EEMD was used to decompose the original vibration signal into several intrinsic mode functions (IMF), the time domain index of the original signal and IMF component were separately calculated; the singular values decomposition for the time domain index were performed to obtain the singular value feature vectors, and the energy ratio of the original signal frequency band and the energy ratio of the IMF component were calculated; IMF component energy ratios, singular value feature vectors and frequency band energy ratios were used as fault feature vectors, which were used as the input to neural network to diagnose and identify the working state of the rotor. The results show that the multidimensional feature vector recognition effect is better than of EEMD energy feature, which can more fully reflect the rotor fault features.