Wear Condition Assessment of Gear Transmission System Based on Wear Debris Boundary Energy

The gear transmission system is the core component in industrial equipment, and its wear state directly affects the reliability and use life of equipment. The wear debris image contains important information on the mechanical wear state. By processing it and analyzing the characteristics and types of wear debris, the health status of mechanical equipment and components can be evaluated. However, wear debris images collected in real time are often affected by Gaussian noise. The improved K-SVD dictionary learning algorithm was used in this paper to remove Gaussian noise, using objective metrics to demonstrate the effectiveness of the improved K-SVD algorithm for wear debris images. Secondly, the improved marked watershed segmentation algorithm (B-FSL) was studied to segment the wear debris chains. After that, the boundary energy (BE) characteristics of the wear debris were extracted to warn about the severe wear state of equipment in advance, an EfficientNetB3 network based on transfer learning was constructed for the recognition and classification of the wear debris image, and the severity of the wear of the mechanical equipment was analyzed. Finally, an experiment was conducted to validate the above methods, proved that the BE characteristics of the wear debris can predict the failure of a planetary gearbox in advance, with the accuracy of the wear debris recognition and classification algorithm exceeding 98%.