As typical failure mode of gear system, the tooth surface pitting, spalling can be detected in most long-running gear system, especially under heavy-load and high-speed condition, or under lubricant-starvation condition, the tooth pitting is characterized by irregular contour and random distribution. Most previous study on the defective gear system mainly based on manually detection of defective region, or just rely on geometric simplification of defects, leading to inaccurate results with low-efficient method, therefore, the machine-vision-based defect inspection method is proposed in the study of defective gear system. First, the pitting defects on gear tooth surface is detected and segmented based on involutional neural network U-net, then the tooth surface with segmented defective region is mapped to the elastohydrodynamic lubrication model of spur gear system, finally, the tribological behavior in addition to the mesh stiffness under lubrication condition of defective spur gear system are investigated and discussed. The results reveal that the machine-vision-based defects inspection could improve the accuracy and efficiency of the failure study for gear system.

中文翻译：

---

使用卷积神经网络自动评估润滑下有缺陷的齿轮系统的啮合刚度

齿面点蚀、剥落是齿轮系统的典型失效形式，大多数长期运行的齿轮系统都会出现齿面点蚀、剥落现象，特别是在重载、高速条件下，或在润滑油匮乏的条件下，齿面点蚀表现为不规则的齿面点蚀、剥落现象。轮廓和随机分布。以往对缺陷齿轮系统的研究大多基于手动检测缺陷区域，或者仅仅依靠缺陷的几何简化，导致结果不准确，方法效率低，因此，提出了基于机器视觉的缺陷检测方法。有缺陷的齿轮系统的研究。首先，基于对合神经网络U-net对齿轮齿面点蚀缺陷进行检测和分割，然后将分割后的缺陷区域的齿面映射到正齿轮系统的弹流润滑模型，最后，除对有缺陷的正齿轮系统在润滑条件下的啮合刚度进行了研究和讨论。结果表明，基于机器视觉的缺陷检测可以提高齿轮系统故障研究的准确性和效率。