A Machine Condition Monitoring Framework Using Compressed Signal Processing
A Machine Condition Monitoring Framework Using Compressed Signal Processing
复制标题
DOI：
10.3390/s20010319
复制
发表时间：
2020-01-01
期刊：
影响因子：
3.9
通讯作者：
Deshmukh, Raghavendra
中科院分区：
文献类型：
--
作者：
Rani, Meenu;Dhok, Sanjay;Deshmukh, Raghavendra
旋转机械滚珠轴承的振动监测是工厂平稳运行和可持续发展的关键方面。使用传统奈奎斯特采样技术的无线振动监测在功耗方面是昂贵的，因为它产生大量需要处理的数据。为了克服这个问题，可以采用压缩感测（CS），其直接获取压缩形式的信号，从而降低功耗。这样生成的压缩测量可以容易地被发送到基站，并且可以在那里使用CS重构算法来恢复原始信号以诊断故障。然而，CS重建在计算时间和功率方面非常昂贵。因此，这种传统的CS框架不适合于真实的实时诊断机械故障。本文提出了一种基于压缩信号处理（CSP）的轴承状态监测框架。CSP是CS的一个较新的研究领域，其中推理问题在不从压缩测量中重建原始信号的情况下得到解决。通过省略重建的努力，所提出的方法显着改善的时间和电力成本。这导致压缩测量的更快处理，以解决机械状态监测所需的推理问题。这为实时诊断机械故障提供了一种方法。所提出的方案与传统方法的比较表明，该方案降低了计算工作量，同时实现可比的故障分类精度。
The vibration monitoring of ball bearings of a rotating machinery is a crucial aspect for smooth functioning and sustainability of plants. The wireless vibration monitoring using conventional Nyquist sampling techniques is costly in terms of power consumption, as it generates lots of data that need to be processed. To overcome this issue, compressive sensing (CS) can be employed, which directly acquires the signal in compressed form and hence reduces power consumption. The compressive measurements so generated can easily be transmitted to the base station and the original signal can be recovered there using CS reconstruction algorithms to diagnose the faults. However, the CS reconstruction is very costly in terms of computational time and power. Hence, this conventional CS framework is not suitable for diagnosing the machinery faults in real time. In this paper, a bearing condition monitoring framework is presented based on compressed signal processing (CSP). The CSP is a newer research area of CS, in which inference problems are solved without reconstructing the original signal back from compressive measurements. By omitting the reconstruction efforts, the proposed method significantly improves the time and power cost. This leads to faster processing of compressive measurements for solving the required inference problems for machinery condition monitoring. This gives a way to diagnose the machinery faults in real-time. A comparison of proposed scheme with the conventional method shows that the proposed scheme lowers the computational efforts while simultaneously achieving the comparable fault classification accuracy.