Search

Diagnosing bearing defects is one of the basic tasks in machine health monitoring, because bearings are critical components of rotating machines. This paper proposes a new method for detecting defects in bearings based on a combination of feature extraction algorithms in which a two-dimensional signal is used. Different from other classical one-dimensional signal processing methods, the proposed method of this paper converts one-dimensional vibration signals into two-dimensional signal (image), then image processing methods are used to analyze the image signal in order to classify the defects that have occurred. arrive at the bearing. Converted images from vibration signals often have specific texture characteristics, and the texture of each defective category is different. In addition, each descriptor extracts spatial features. Some features are weak and others are strong. In this article, the method of removing additional key points of SIFT (RKEM SIFT) is used. In addition, for each descriptor, the best features are selected using the non-linear principal component analysis method. Finally, the selected features are combined and four classification methods are applied to achieve the best classification performance and after comparison, the best classification method is selected. The performance of the proposed algorithm is evaluated on the standard bearing data set of Case Western Reserve University. The simulation results show that the proposed method performs better than other methods of fault finding of rolling bearings.