A Fault Diagnosis Method Based on EMD and PCA

In order to enhance fault diagnosis accuracy while minimizing computational complexity, this study presents a fault feature extraction method based on EMD (Empirical Mode Decomposition) and PCA (Principal Component Analysis). The EMD is employed to decompose the original signal into a set of Intrinsic Mode Functions (IMFs). The energy of each IMF and its ratio to the energy of the original signal are computed. The dominant IMFs are selected based on their energy ratio, and their energies are utilized as partial fault features. Additionally, time-domain indicators of the original signal are calculated, and the principal components are extracted as additional fault features using PCA based on their contribution rates. Finally, the two sets of fault features are combined to obtain the primary fault features, which serve as inputs for a BP neural network to classify fault states. The proposed method is evaluated using a bearing fault dataset. The results demonstrate that the root mean square error of the fault diagnosis results of the primary fault features is 0.0743, which is lower than that of original fault features (0.1263). The proposed method effectively extracts key fault feature information, leading to improved diagnosis accuracy and reduced computational complexity.