Article
Compressed-Sensing Reconstruction Based on Block
Sparse Bayesian Learning in
Bearing-Condition Monitoring
Jiedi Sun
1,
*, Yang Yu
2
and Jiangtao Wen
2
1
School of Information Science and Engineering, Yanshan University, 438, Hebei Avenue,
Qinhuangdao 066004, China
2
Key Laboratory of Measurement Technology and Instrumentation of Hebei Province, Yanshan University,
Qinhuangdao 066004, China; paulyu2016@163.com (Y.Y.); wens2002@163.com (J.W.)
* Correspondence: sjdwjt@ysu.edu.cn; Tel.: +86-335-805-7078
Received: 24 March 2017; Accepted: 15 June 2017; Published: 21 June 2017
Abstract:
Remote monitoring of bearing conditions, using wireless sensor network (WSN),
is a developing
trend in the industrial field. In complicated industrial environments, WSN face
three main constraints: low energy, less memory, and low operational capability. Conventional
data-compression methods, which concentrate on data compression only, cannot overcome these
limitations. Aiming at these problems, this paper proposed a compressed data acquisition and
reconstruction scheme based on Compressed Sensing (CS) which is a novel signal-processing
technique and applied it for bearing conditions monitoring via WSN. The compressed data acquisition
is realized by projection transformation and can greatly reduce the data volume, which needs the
nodes to process and transmit. The reconstruction of original signals is achieved in the host computer
by complicated algorithms. The bearing vibration signals not only exhibit the sparsity property,
but also have specific structures. This paper introduced the block sparse Bayesian learning (BSBL)
algorithm which works by utilizing the block property and inherent structures of signals to reconstruct
CS sparsity coefficients of transform domains and further recover the original signals. By using the
BSBL, CS reconstruction can be improved remarkably. Experiments and analyses showed that BSBL
method has good performance and is suitable for practical bearing-condition monitoring.
Keywords:
compressed sensing reconstruction; sparse Bayesian learning; block sparse structure;
bearing condition monitoring; wireless sensor network
1. Introduction
As critical components in rotating machinery, bearings that are not in a good condition can
cause frequent machinery breakdowns [
1
], and these faults may result in equipment instability,
poor efficiency, and even major production-safety accidents [
2
]. A stable machine-condition monitoring
(MCM) system is required to guarantee the optimal states of the bearings during operation [
3
]. Various
physical properties can be utilized to monitor and diagnose the bearing faults. Most MCM systems in
the industrial field are based on vibration signals, which are easy to acquire and can provide complete
information [4].
In modern industries, some problems exist in online wired MCM systems, such as installation
difficulty and high cost, limited power supply and additional long cables. The wireless sensor network
(WSN) offers a novel approach to improve the traditional wired MCM systems [
5
], and it has some
advantages such as rapid deployment, removability, and low energy consumption [
6
]. However,
the WSN manifests a number of limitations when applied in vibration-based MCM systems. According
to the Nyquist–Shannon sampling theorem, an analog-to-digital converter (ADC) in WSN nodes
Sensors 2017, 17, 1454; doi:10.3390/s17061454 www.mdpi.com/journal/sensors
