目的通过构建轮毂在线生产视觉检测系统,预测轮毂生产过程中轮毂表面的缺陷。方法根据轮毂表面缺陷的定义和评价标准,给出了轮毂表面缺陷的计算模型,采用了改进型的Faster-RCNN目标检测算法,引入了深度生成式对抗网络,消除图像的模糊性...目的通过构建轮毂在线生产视觉检测系统,预测轮毂生产过程中轮毂表面的缺陷。方法根据轮毂表面缺陷的定义和评价标准,给出了轮毂表面缺陷的计算模型,采用了改进型的Faster-RCNN目标检测算法,引入了深度生成式对抗网络,消除图像的模糊性,再利用清晰的轮毂表面图像进行模型训练,结合领域专家的判别标准,优化网络参数,构建轮毂表面缺陷检测模型。利用深度学习Pytorch框架,在NVIDIA Tesla P100图像加速卡上进行模型训练,并对模型结果进行对比性实验分析,找出最优的预测模型。结果在基础网络部分,采用残差模型ResNet101网络比采用VGG16模型的准确率提高了24%。在目标检测网络模型中引入了多通道特征融合模块,准确率提升了2%。再引入FPN金字塔模型,融入低级和高级语义信息,使得输出的多尺度的预测特征图谱效果更好。最后把残差网络的ROI-Pooling算法改为ROI-Align算法,准确率提高了5%。通过对网络模型的不断改进和优化,轮毂表面缺陷的识别率不断提高。结论利用改进型的Faster-RCNN网络能够识别出轮毂表面缺陷的种类和位置,满足生产环境的要求,具有一定的工程应用价值。展开更多
The recent research and development of forged magnesium road wheel were reviewed.Methods of flow-forming,spin forging of manufacturing a forged magnesium alloy wheel were introduced.A new extrusion method was investig...The recent research and development of forged magnesium road wheel were reviewed.Methods of flow-forming,spin forging of manufacturing a forged magnesium alloy wheel were introduced.A new extrusion method was investigated especially. Extrusion from hollow billet was proposed in order to enhance the strength of spoke portion and reduce the maximum forming load. By means of the developed technique,the one-piece Mg wheels were produced successfully by extrusion from AZ80+alloy.At the same time,the existing problems on the research and development of forged magnesium road wheel were analyzed.The impact testing,radial fatigue testing and bending fatigue testing results show that AZ80+wheel can meet application requirement in automobile industry.展开更多
文摘目的通过构建轮毂在线生产视觉检测系统,预测轮毂生产过程中轮毂表面的缺陷。方法根据轮毂表面缺陷的定义和评价标准,给出了轮毂表面缺陷的计算模型,采用了改进型的Faster-RCNN目标检测算法,引入了深度生成式对抗网络,消除图像的模糊性,再利用清晰的轮毂表面图像进行模型训练,结合领域专家的判别标准,优化网络参数,构建轮毂表面缺陷检测模型。利用深度学习Pytorch框架,在NVIDIA Tesla P100图像加速卡上进行模型训练,并对模型结果进行对比性实验分析,找出最优的预测模型。结果在基础网络部分,采用残差模型ResNet101网络比采用VGG16模型的准确率提高了24%。在目标检测网络模型中引入了多通道特征融合模块,准确率提升了2%。再引入FPN金字塔模型,融入低级和高级语义信息,使得输出的多尺度的预测特征图谱效果更好。最后把残差网络的ROI-Pooling算法改为ROI-Align算法,准确率提高了5%。通过对网络模型的不断改进和优化,轮毂表面缺陷的识别率不断提高。结论利用改进型的Faster-RCNN网络能够识别出轮毂表面缺陷的种类和位置,满足生产环境的要求,具有一定的工程应用价值。
基金Project(50735005)supported by the National Natural Science Foundation of China
文摘The recent research and development of forged magnesium road wheel were reviewed.Methods of flow-forming,spin forging of manufacturing a forged magnesium alloy wheel were introduced.A new extrusion method was investigated especially. Extrusion from hollow billet was proposed in order to enhance the strength of spoke portion and reduce the maximum forming load. By means of the developed technique,the one-piece Mg wheels were produced successfully by extrusion from AZ80+alloy.At the same time,the existing problems on the research and development of forged magnesium road wheel were analyzed.The impact testing,radial fatigue testing and bending fatigue testing results show that AZ80+wheel can meet application requirement in automobile industry.
