An Intelligent Analysis Method for Ferrography Images and Its Applications

Ferrography is widely used for equipment condition monitoring, with wear particle coverage serving as an important indicator of wear conditions. However, ferrography images often suffer from complex backgrounds prone to segmentation failures and contain non-wear contaminants, which can interfere with the calculation. To address these issues, we propose an intelligent ferrography image analysis method to calculate wear particle coverage. An image segmentation evaluation metric (ISEM) is developed to select a suitable binary mask from five threshold segmentation methods without requiring ground-truth masks. A YOLOv12 model incorporating Poly Kernel Inception (PKI) then detects contaminants for removal to obtain corrected wear particle coverage. The proposed method was applied to two datasets comprising 234 ferrography images from the testing of an aero-engine accessory gearbox and the accelerated fatigue testing of accessory bearings. On a manually annotated subset of 80 gearbox images, ISEM selected a method attaining the highest precision among the five candidates in 97.5% of images. The contaminant detection model achieves an mAP50 of 82.42 ± 1.30% on the test set, while contaminant removal reduces MAE from 1.06% to 0.29% on the annotated subset. These results demonstrate improved accuracy of wear particle coverage calculation, providing a quantitative basis for equipment condition monitoring.