Transfer Learning-Based Fault Diagnosis of Single-Stage Single-Acting Air Compressor
参考中译:基于传递学习的单级单作用空气压缩机故障诊断


          

刊名：Journal of Vibration Engineering & Technologies
作者：G. Chakrapani（School of Mechanical Engineering (SMEC), Vellore Institute of Technology）
S. Naveen Venkatesh（School of Mechanical Engineering (SMEC), Vellore Institute of Technology）
S. Aravinth（School of Mechanical Engineering (SMEC), Vellore Institute of Technology）
V. Sugumaran（School of Mechanical Engineering (SMEC), Vellore Institute of Technology）
刊号：712HA010
ISSN：2523-3920
出版年：2024
年卷期：2024, vol.12, no.3 Pt.2
页码：4411-4428
总页数：18
分类号：TB1
关键词：Air compressor；Deep learning；Pre-trained models；AlexNet；GoogLeNet；ResNet50；VGG19；VGG16
参考中译：空气压缩机;深度学习;预训练模型; AlexNet; GoogLeNet; ResNet 50; VGG 19; VGG 16
语种：eng
文摘：Introduction Reciprocating air compressor, which is also known as piston compressor is one of the crucial machinery used in various production lines to move gas at high pressure. Research question The prolonged operation of this machine can lead to internal damage. Therefore, it is highly important to incorporate fault diagnosis to prevent sudden and unforeseen failure. Condition monitoring and fault diagnosis of machines are becoming more and more crucial in various industries. They have special importance in places where the breakdown of machines can cause a tremendous financial crisis. Although there are extensive research works in this area, fault diagnosis of reciprocating air compressors using deep learning is still unexplored. Methodology In this paper, the condition monitoring of air compressors is discussed using deep learning methods. First, different modes of signal acquisition are surveyed, and the best one among them is chosen. Out of several faults, five significant faults that are prone to air compressors are taken into study. Magnitudes of vibration signals are captured using the accelerometer sensor. These signals are converted to plots using MATLAB and the faults are classified using pretrained networks like AlexNet, GoogLeNet, ResNet50, VGG19 and VGG16. Results The results obtained show that the AlexNet pretrained network exhibits the best fault classification rate of 100% in a minimum computational time of 570 s.
参考中译：介绍往复式空压机，又称活塞式空压机，是各种生产线上用于高压输送气体的关键机械之一。研究问题这台机器长时间运行会导致内部损坏。因此，结合故障诊断以防止突发和不可预见的故障是非常重要的。机械设备的状态监测和故障诊断在各个行业中变得越来越重要。在机器故障可能导致巨大金融危机的地方，它们具有特别重要的意义。虽然在这方面已经有了大量的研究工作，但基于深度学习的往复式空压机故障诊断还没有得到深入的探索。本文采用深度学习方法对空压机的状态监测进行了研究。首先，对不同的信号采集方式进行了比较，从中选出了最优的一种。在几个故障中，研究了空气压缩机易发生的五个重大故障。使用加速度计传感器捕获振动信号的大小。这些信号使用MatLab转换成曲线图，并使用AlexNet、GoogLeNet、ResNet50、VGG19和VGG16等预先训练的网络对故障进行分类。结果在最小计算时间为570 S的情况下，AlexNet网络的故障分类效果最好，分类正确率为100%。



国家科技图书文献中心  全球文献资源网  京ICP备05055788号-26  京公网安备11010202008970号  机械工业信息研究院 2018-2024