A gear gluing degree measurement method based on visual segmentation

By using the improved U-Net semantic segmentation network DSA-Net, combined with reflection-invariant enhanced convolution and position-aware dynamic sampling module, the problem of detecting gear scuffing defects under complex lighting conditions was solved. This enabled accurate quantification of high-reflectivity interference and irregular boundaries on the metal tooth surface, improving detection efficiency and accuracy.

CN122265199APending Publication Date: 2026-06-23CHONGQING UNIV
View PDF -1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHONGQING UNIV
Filing Date
2026-03-23
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately identify and quantify gear scuffing defects under complex lighting conditions, particularly in adaptive modeling of high reflectivity interference from metal tooth surfaces and irregular scuffing boundaries, leading to low detection efficiency and insufficient accuracy.

Method used

An improved U-Net semantic segmentation network, DSA-Net, is adopted, combined with the reflection-invariant enhanced convolutional module DSA-DSConv and the position-aware dynamic sampling module. Through data augmentation processing of training samples, high reflection interference of metal tooth surface is suppressed and irregular glued boundary is adaptively modeled to achieve accurate quantization of glued region.

Benefits of technology

It effectively suppresses high reflection interference from metal tooth surfaces, adaptively models irregular bonding boundaries, achieves accurate quantification of the bonding area, provides objective and quantifiable evaluation indicators, and improves the accuracy and reliability of detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122265199A_ABST
    Figure CN122265199A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of mechanical fault diagnosis and intelligent visual inspection technology, and particularly relates to a method for measuring the degree of gear scuffing based on visual segmentation. The method includes: S1, acquiring images of the scuffed surface of the gear under complex lighting and high reflectivity conditions; S2, generating a training set; S3, constructing an improved U-Net semantic segmentation network DSA-Net, where the encoder of DSA-Net consists of stacked reflection-invariant enhanced convolutional modules DSA-DSConv, and the decoder includes a position-aware dynamic sampling module; S4, training the DSA-Net model to obtain the optimal segmentation model; S5, inputting the scuffed surface image into the optimal segmentation model to output a segmentation mask image of the scuffing defect; S6, post-processing the segmentation mask image and calculating the ratio of the pixel area of ​​the final scuffed region mask to the total pixel area of ​​the effective tooth surface region mask, using this ratio as a quantitative evaluation index of the degree of gear scuffing. This method can effectively suppress high reflectivity interference from metal tooth surfaces, adaptively model irregular scuffing boundaries, and achieve accurate quantification of the scuffed region area.
Need to check novelty before this filing date? Find Prior Art