基于AlexNet和迁移学习的滚动轴承故障诊断研究

Fault diagnosis of rolling bearing based on AlexNet and transfer learning

传统的轴承故障诊断方法需要进行复杂的信号处理,同时依赖专家知识和人工构造算法等技术手段,并且工程实际中可利用的机械设备故障数据量较少,针对这一系列问题,以滚动轴承正常运行时和发生不同故障时收集到的原始振动信号为识别依据,提出了一种基于AlexNet和迁移学习的滚动轴承故障诊断方法。将收集到的滚动轴承原始振动数据转换为振动信号图,并为振动信号图设定了标签,以作为训练样本;对预训练的AlexNet网络进行了微调,以使其符合任务需求,并使用准备好的训练样本对网络进行了训练;使用美国凯斯西储大学轴承数据中心的数据集,对网络模型的性能进行了验证,在滚动轴承的内圈故障、外圈故障和滚动体故障3个故障类别下,达到了100%的诊断精度。研究结果表明:在标记故障数据稀缺的情况下,采用该方法仍可实现对滚动轴承常见故障类型的诊断,且与现有先进方法相比,该方法的诊断精度有所提升。

Aiming at the problems that complex signal processing,expert knowledge and artificial construction algorithms and other technical means were needed in traditional bearing fault diagnosis methods and few fault data of mechanical equipment was available in engineering practice,a rolling bearing fault diagnosis method based on AlexNet and Transfer Learning was put forward by taking the original vibration signal of rolling bearing in normal operation and different faults as identification basis.The collected original vibration data of rolling bearings were converted into vibration signal diagrams,and then the labels for the vibration signal diagrams were set as training samples.The pre-trained AlexNet network was fine-tuned to meet the task requirements,and the prepared training samples were used to train the adjusted network.The data set of the Bearing Data Center of Case Western Reserve University in the United States was used to verify the performance of the network model,and the diagnosis accuracy of 100%under three categories,inner ring fault,outer ring fault and rolling element fault,was obtained.The research results show that this method can realize the diagnosis of common fault types of rolling bearings even when the marked fault data is scarce,and the diagnosis accuracy improves comparing with existing advanced methods.