Extraction of Information From Vibration Data Using Double Density Discrete Wavelet Analysis for Condition Monitoring

Many condition monitoring (CM) techniques have been investigated for the purpose of early fault detection and diagnosis in order to avoid unexpected machine breakdowns. However, non-stationary and non-linear characteristics of vibration data can make the signal analysis a challenging task. Multiresolution data analysis approaches have received significant attention in recent years and are widely applied to analyse non-stationary and non-linear data. Double-Density Discrete Wavelet Transform (DD-DWT), which was originally developed for image processing, is proposed and investigated in this paper for effectively extracting diagnostic features from the vibration measurements. DD-DWT has the merits of nearly shift-invariant and less frequency aliasing which and allows the effective extraction of non-stationary periodic peaks, compared with the undecimated DWT. Techniques based on thresholding of wavelet coefficients are gaining popularity for denoising data. The implementation of global, level-dependent, and subband-dependent thresholding based methods are investigated and implemented on the selected wavelet coefficients in order to denoise and enhance the periodic and impulsive fault features. The performance of the proposed method has been evaluated against DWT using both simulated data and experimental datasets from defective tapered roller bearings. Results, using the harmonic to signal ratio (HSR) as a measure, have demonstrated that DD-DWT outperforms conventional DWT in feature extraction and noise suppression. As a result, the proposed method is robust and effective in fault detection and diagnosis.