Image Denoising and Feature Extraction of Wear Debris for Online Monitoring of Planetary Gearboxes

Wear debris generated in a pitting process of planetary gearboxes carries valuable information about health status. However, RGB images of online wear debris are often affected by image blur caused by Gaussian noise, high-frequency noise, and other random noises besides Gaussian noise, including bubbles in lubricating oil, dark oil caused by contamination, and the temperature rise of electronic components. To address these issues, in this work, an image denoising model WVBOD was proposed based on the fusion of wavelet, variational mode decomposition and non-local mean filtering, which makes full use of the advantages of above three denoising methods, removes the noise in the image and preserves the details of the image information. Comparing the peak signal-to-noise ratio and structural similarity of the denoised image using different models, the WVBOD objectively acquired better denoising result than other advanced denoising methods. In addition, the previous online wear debris features mainly focus on using changes in particle concentration to reveal the deterioration of the wear state. Based on the fact that the quantity of large wear debris increases with the evolution of gear pitting, a novel wear index
$Z(i)$
, representing the size gradient of large wear debris and sensitive to an increase in large wear debris, is proposed for the denoising image. Then early fault warning can be realized for the planetary gearbox. Finally, it is verified by the size and quantity features extracted by using offline oil analysis techniques.