Investigating of a sensitive intelligent method for damage identification of roller bearing under various external conditions

Roller bearings are widely used in rotating machinery and are very important so that one of the major reasons for machine breakdown is their failure. Although numerous studies have been done for damage identification using empirical mode decomposition (EMD) and feature extraction from intrinsic mode functions (IMFs), EMD has some drawbacks such as sifting stop criterion and mode mixing which make some limitations. In addition, acquired vibration signals used to damage identification are usually noisy which have great influence on the results. Thus, one of the great challenges of EMD technique is that, especially for small defect sizes, it seems not to be able to recognize healthy and faulty condition of a bearing. An intelligent sensitive method is needed to detect damage automatically before growing the defect size. Ensemble empirical mode decomposition (EEMD) is a newly developed noise assisted method to solve mode mixing problem which is a consequence of signal intermittence. On the other hand, Wavelet packet decomposition (WPD) is a powerful technique used for denoising of acquired signal effectively. In this study it is investigated to identify a sensitive automatic method which is able to diagnose very small defects of roller bearings under various working conditions (load and speed) through processing of the acquired signals; the latest are generated by means a test rig assembled by dynamics & identification research group (DIRG) at mechanical and aerospace engineering department , Politecnico di Torino. Then the accuracy and sensitivity of the method is examined for external effects removal by classifying the samples of different external conditions whether they are healthy or faulty using support vector machine (SVM