Extreme learning machine based on stationary wavelet singular values for bearing failure diagnosis

Intelligence condition monitoring based on vibration signal analysis plays a key role in improving rolling bearings failure diagnosis in mechanical transmission systems. Unexpected failures in the bearings may cause machine breakdowns that are very expensive for the industry. Hence, this study proposes a method to the rolling element bearing failure diagnosis which is based on an extreme learning machine (ELM) algorithm combined with stationary wavelet transform (SWT) and singular value decomposition (SVD). The SWT is used to separate the vibration signals into a series of wavelet component signals. Then, the obtained wavelet components matrix is decomposed by means of a SVD method to obtain a set of wavelet singular values. Finally, the wavelet singular values are used as input to the extreme learning machine for classification among ten different bearing failure types. Obtained results using the proposed model shown high accuracy of diagnosis under variable speed condition.