Abstract: Existing bearing fault diagnosis technology still has the following two problems:First,features need to be extracted manually in traditional diagnostic methods,which is time-consuming and unstable;Second,huge amounts of computing resources and time are needed in Convolutional Neural Networks(CNN)model training,which can not fulfill the real-time response requirements of fault diagnosis.To solve above problems,a cloud/edge collaborated fault diagnosis solution with real-time response ability for bearing was proposed.The experiment showed that this method had higher accuracy and lower time cost than the method without cloud/edge cooperation when the sample amount was small.The improved bearing fault diagnosis algorithm achieved higher diagnosis accuracy.Through transfer learning of the improved algorithm and edge-side collaboration,individualized product adaptability and real-time diagnosis ability of the fault diagnosis algorithm were enhanced.

Key words: intelligent fault diagnosis, cloud/edge collaboration, convolutional neural networks, transfer learning

摘要： 现有轴承故障诊断技术存在以下问题:①传统诊断方法需要人工提取特征,耗时长,诊断结果不稳定;②卷积神经网络诊断方法需要大量的计算资源和较长的训练时间,与故障诊断的实时响应要求存在矛盾。针对以上问题,提出一种云/边缘协同的实时轴承故障诊断方案。经过实验验证,该方案在拥有少量样本情况下与不进行云/边缘协同相比可大幅提高诊断准确率,并节约大量的训练时间。通过改进的轴承故障诊断算法达到了较高的故障诊断准确性,并通过模型的迁移学习与边缘端协同,增强了故障诊断算法对个性化应用的适应性和故障诊断的实时性。

关键词: 智能故障诊断, 云/边缘协同, 卷积神经网络, 迁移学习