基于深度学习的目标部件故障识别技术及应用

Automated visual inspection of target parts based on deep learning

图像识别是故障检测的一种常见方法,传统的图像识别技术需要特定算法提取图像特征,然而特征提取依赖大量可观察的显著图像信息,对于某些信息量不足易受环境干扰的微小部件,识别精度往往较低。对此提出了基于深度学习的目标部件识别方法,将传统图像算法与深度学习相结合,有效提升了识别效率。在堆叠自编码器(SAE)预训练的基础上,建立了卷积神经网络(CNN),以货车故障动态图像检测系统(TFDS)为应用背景,研究了货车心盘螺栓的丢失故障检测。最终识别精度高于98%,具有较高的工程应用前景。

Image recognition is a common method of visual inspection. Traditional visual inspection relying on feature extrac-tion cannot always meet the requirements of high-accuracy inspection, for some key parts, such as fastening bolts, do not possess sufficient feature information. To resolve the issues mentioned above, a method that combines tradi-tional visual inspection with deep learning was proposed. Traditional feature extraction is used to locate the targets approximately, then a composite neural network, SAE-CNN, is provided to further improve the training efficiency. A stacked auto-encoder(SAE) is added to a convolutional neural network(CNN) so that the network can obtain optimum results faster and more accurately. Taking the inspection of center plate bolts in a moving freight car as an example, the overall system and specific processes are described. Finally, the result with 98% accuracy is showed integrally.