Research On Rotating Machinery Vibration Signal Processing Method Based On Compressed Sensing

The machinery condition monitoring system based on wireless sensor network(WSN)has several advantages,such as low installation cost,simple layout method,and strong expansibility.However the sensor nodes have low operational capabilities,small memory and limited power supply,which brings some difficulties for the WSN application in machinery vibration signal processing.It is significant to decrease the data size before wireless transmission,in order to decrease nodes power dissipation and increase node lifetimeInformation sampling method based on Compressed Sensing(CS)achieves the synchronization of data acquisition and compression.This paper researches to CS key problems for rotating machinery vibration signal process,which include reconstruction algorithms,sparse representation and measurement matrix.Aim at the property of rotating machinery vibration signal,this paper improves the traditional CS framework,which supply a new method to low power consumption data acquisition in WSN nodes.The specific research contents are as follow.Firstly,this paper study the block sparse property of bearing vibration signals,and propose that using block sparse Bayesian learning(BSBL)framework as CS reconstruction algorithm.The reconstruction performance is compared with traditional reconstruction algorithms through experiments.Next,based on multiresolution analysis,this paper builds the wavelet packet transformation(WPT)matrix.Then,combining WPT matrix and K-SVD dictionary training method,this paper proposes a new sparse representation dictionary construction method baseed wavelet packet dictionary optimization(WPDO),which is compared with orthogonal basis sparse representation dictionary through experiments.Finally,aiming at the implementation difficulties of random measurement matrix,this paper research an information acquisition method via random demodulation(RD),and improves the parameters of filter parameters,and proposed a certain measurement matrix construction method based on RD and improved Low Density Parity Check Code(LDPC)party-check matrix.The feasibility of this method is verified through experiments.