A method for size estimation of bearing faults from vibration signals based on denoising, envelope and WPD

Abstract An efficient fault prognosis for bearings remains still as an open challenge within the industry. Bearing damage is primarily derived from wear or surface fatigue in rotational machinery. This damage especially at early stages , is difficult to characterize due to the weakness of the vibration signal and external interferences, which is why advanced signal processing is required. To tackle this challenge, this study introduces the DEW method, a combination of Wavelet Denoising, Envelope, and Wavelet Packet Decomposition, to estimate the size of a bearing fault located in inner and/or outer races. This method is validated with a dataset of varying rotational speeds, loads, and fault sizes obtained from open-access sources. The application of the method enabled the estimation of a confidence interval for fault size under different conditions. Despite some observed inconsistencies, the method showed more accurate results with increased load, fault size, or decreased speed. In the case of simulated signals , good results were achieved even in scenarios with up to 10 [dB] of signal-to-noise ratio. However, for the evaluation with the dataset, this estimation proved insufficient due to the amount of noise present, which attenuated the components of interest, especially the input peak.