Abstract:Aiming at the current situation that the labeled vibration data of rolling bearings is small and the failure modes are diverse, a semi-supervised fault diagnosis method for bearing vibration signal is proposed, which is based on AFSA-FCM-ISODATA (AFI) hybrid clustering algorithm. The wavelet packet decomposition method is used to extract the signal’s energy feature spectrum, and the principal component analysis method is used to enhance the feature. Referring to the idea of "split" and "merge" in iterative self-organizing data analysis, we add "split evolution" and "merge evolution" behaviors for individual fish in the artificial fish swarm algorithm. Using the membership matrix and loss function defined by the fuzzy C-means method and the improved artificial fish swarm algorithm, the global optimal solution of the objective function can be searched, which means the centers of clusters representing each failure mode are also found. Fault diagnosis then can be realized by calculating the nearest neighbor cluster center of the test data. The results show that this method not only does not need to specify the number of clusters, but also can complete the training with a small amount of labeled data. Compared with other similar methods, it also achieves better failure modes recognition performance.