Bearing fault diagnosis based on feature fusion

Locally linear embedding algorithm is widely utilized to feature extraction, and it is also an effective tool for fault diagnosis. However, locally linear embedding is sensitive to redundancy dimensions, that is, too high dimensions may degrade the performance of it. In order to address this problem, in this paper, a novel feature fusion method is proposed, which integrates a feature selection algorithm into locally linear embedding. In this paper, we first employ Fisher criterion to roughly select the representative features from original data set, and then locally linear embedding is used to extract the final low-dimensional results. Experiments results show that the proposed method is robust to the value of neighbors and can obtain better classification accuracy.