Analysis And Research On The Reconstruction Algorithm Of Bearing Fault Vibration Signal Based On Compressive Sensing

Rolling bearings,as an important component supporting the mobility of equipment,often cause equipment abnormalities due to their failures.However,in traditional rolling bearing fault detection based on bearing vibration signals,the high sampling frequency during rolling bearing operation brings a large amount of data,which causes great pressure on real-time remote signal transmission or storage systems.Compressive sensing,as an emerging data compression sensing technology solution,integrates traditional data acquisition and compression.It does not require complex data coding processing algorithms to achieve synchronous compression sampling of signals.The amount of compressed signal data obtained is small enough,but contains enough The amount of original signal characteristic information.Therefore,the construction of a bearing vibration signal compression sensing frame to achieve bearing vibration signal compression will effectively alleviate the related problems caused by the huge amount of sampled data.Based on this,this dissertation expands the research on related technologies in the bearing vibration signal compression sensing framework,optimizes the sparse representation method of bearing vibration signals,and proposes two improvements to the compressed sensing reconstruction algorithm;in order to verify the accuracy of the results achieved by the framework,An improved algorithm of the traditional threshold function is proposed,and the actual fault vibration signal is used for frame verification test under different compression ratios.The main research contents are as follows:1)In order to improve the traditional greedy algorithm reconstruction time cost and poor reconstruction performance and other problems,the cosine generalized orthogonal matching pursuit(Cosine Generalizsd Orthogonal Matching Pursuit,CGOMP)algorithm is proposed.By adding the cosine similarity coefficient optimization strategy,the utilization of data atoms is improved,and the search of the best candidate atoms is strengthened.By comparing with existing greedy algorithm performance experiments,the algorithm proposed in this paper has great advantages in average support set error,accurate reconstruction rate,and relative error performance.2)Aiming at the problem of insufficient reconstruction accuracy a nd stability in the convex optimization algorithm,a modified Newton Smoothing Norm(Optimal Newton Smooth L0 Norm,ONSL0)algorithm under the optimal control strategy is proposed.This algorithm not only optimizes the arc sine function with a high degree of approximation to the discrete L0 norm,but also in order to avoid "jaggy" in the iterative direction,the modified damping Newton method is used to control the iterative solution direction,and the maximum number of internal loops is controlled.In order to further improve the iterative solution efficiency of the algorithm,the decrement parameter is modified by the attenuation control strategy.Through experimental comparison and analysis,the algorithm is better than similar algorithms in reconstruction signal-to-noise ratio,reconstruction time cost and reconstruction accuracy,which confirms the effectiveness of this algorithm.3)Aiming at the analysis and diagnosis of the processing results of the bearing vibration signal compression sensing frame,an improved wavelet threshold fault characteristic frequency identification method is proposed.The threshold function can flexibly realize the mode approximation of soft and hard thresholds.Experiments show that this scheme not only has advantages in signal-to-noise ratio and mean square error,but also enhances the smoothness of signal processing.It can reduce the noise of the bearing vibration signal,and combined with the power spectrum can effectively identify the fault characteristic frequency.