Abstract: Traditional machine learning algorithms used for fault diagnosis in the automated production lines of automobiles require that the training and test sets have the same distribution and need a substantial number of training samples. However, in practice, fault sample data are difficult to acquire, and the operating conditions of production lines are highly variable, leading to a low fault classification accuracy. In view of these problems, this paper proposes a research method for fault diagnosis in automated automobile production lines based on Digital Twin (DT) technology. This method initially models the actual production lines using SolidWorks, followed by rendering through Unity 3D software, and combines with PLC for DT model simulation. Finally, the method utilizes transfer learning techniques and convolutional neural networks to achieve fault diagnosis. The feasibility of the proposed method is verified by comparison with existing methods.

Key words: DT, fault diagnosis, unity3D, automated production line, convolutional neural network

摘要： 传统的机器学习算法对汽车自动化生产线开展故障诊断研究,需满足训练集和测试集具有相同的分布,且需要较多的训练样本,但在实际中,故障样本数据难以获取、生产线运行工况多变,导致故障分类准确率较低。鉴于以上问题,文中提出了一种基于数字孪生(Digital Twin,DT)技术的汽车自动化生产线故障诊断研究方法,该方法首先使用SolidWorks对实际生产线建模,然后通过Unity3D软件进行渲染,并结合PLC进行DT模型仿真。最后结合迁移学习技术和卷积神经网络技术实现故障诊断,并与现有方法进行了对比,验证了所提方法的可行性。

关键词: DT, 故障诊断, Unity3D, 自动化生产线, 卷积神经网络