Gearbox Fault Diagnosis Based on a Novel Hybrid Feature Reduction Method

The dimensionality reduction of the high-dimensional feature space is a critical part for data preprocessing, which directly affects the accuracy of fault diagnosis. In this paper, a novel hybrid algorithm named principal component locally linear embedding (PCLLE) is introduced to compress the original high-dimensional feature. This approach combines the optimization objectives of the principal component analysis (PCA) and locally linear embedding (LLE), which attempts to find a mapping that meets the optimization goals of PCA and LLE at the same time. It is applied on the gearbox fault diagnosis. In the experiment, the extracted fault-sensitive feature is compressed by PCLLE method. Then, the compressed feature is embedded with five classifiers for fault detection. To evaluate the performance of the proposed new method, the traditional PCA and LLE methods are introduced for comparison. Experimental results show that the PCLLE algorithm has good performance during the classification process compared with the traditional PCA and LLE method.