Research On Joint Sparse Reconstruction Algorithm Of Mechanical Vibration Signals Under Multiple Monitoring Points

The mechanical equipment generated vibration signal under the running state,which contains a rich of information.Therefor,on-line monitoring and real-time ac-quisition in this state to obtaining status through fault diagnosis is critical to improv-ing mechanical equipment performance,ensuring product quality,reducing mainte-nance costs,and improving business efficiency.However,the mechanical equipment has a inevitable trend,which is high speed,integration,intelligence and complexity.At this time,the frequency of mechanical vibration signals is higher,and a huge amount of data will be generated when the traditional Nyquist sampling law is still used for sampling,the processing and storage of these data has become an urgent problem to be solved,and the emergence of compressed sensing has made a qualita-tive leap to solve this problem.The traditional compressed sensing is single meas-urement vectors model,which obtain a single measurement vector under single point measurement conditions,but the compressed sensing under multiple monitoring points is multiple measurement vectors model.Compared with the SMV model,the MMV model can further utilize the correlation between the data that is more condu-cive to accurately estimating the position of the non-zero rows accurately,thus ob-taining a more accurate estimation solution.Reconstruction algorithm as an essential part whether compressive sensing theory can be applied to practical engineering,in this paper,compressive sensing is applied to the acquisition of mechanical vibration signals,and the joint sparse reconstruction algorithm of mechanical vibration signals under multiple measurement vectors model is studied.The main research results of the thesis are as follows:?1?The basic theory and model of compressed sensing are introduced.It mainly includes three aspects:sparse representation theory,measurement matrix design,re-construction algorithm research.In addition,the reconstruction principles and com-mon reconstruction methods of three different models,SMV,MMV,Infinite Meas-urement Vectors?IMV?are introduced.The reconstruction algorithm of multiple measurement vector models is described by using the classical matching algorithm of greedy algorithm,the reconstruction performance in different measurement variables is analyzed by simulation experiments.?2?The sparse Bayesian algorithm for mechanical vibration signals under multi-ple measurement vectors is studied.When the column of the perceptual matrix have a highly correlation,most of the reconstruction algorithms have a poor performance?such as ₁l algorithm,matching pursuit algorithm and orthogonal matching pursuit algorithm?,but the sparse Bayesian algorithm is still well.What's more,the sparse Bayesian algorithm has a pretty recovery effect when it solves the signal with strong inter-and intra-signals correlation and time structure correlation.Under the condition of multiple monitoring points,the mechanical vibration signals are MMV model with strong temporal structure correlation.Based on this inherent feature of mechanical vibration signals,the basic principle of sparse Bayesian algorithm with multiple measurement vector models and the sparse Bayesian algorithm with timing structure are studied and analyzed.Meanwhile,the simulation was carried out and the results show that,the two sparse Bayesian algorithms have better recovery effect,the recon-struction error is small and has good adaptability for the mechanical vibration signal under the MMV model.?3?A joint sparse reconstruction algorithm of mechanical vibration signals based on particle swarm optimization is proposed.Different from the traditional algorithms,particle swarm optimization is an effective modern intelligent method for solving combinatorial optimization problems.Aiming at the problems,such as the recovery performance of the traditional greedy algorithm is poor,the recovery efficiency of the convex relaxation algorithm is low,the randomization of the population initialization of the standard particle swarm algorithm is too strong,fall into precocity easily and enter the local search in the process of position updating,a joint sparse reconstruction method of mechanical vibration signals based on particle swarm optimization is pro-posed.Firstly,the correlation sparse Bayesian algorithm is used as the initial solution.Then the least square method is used to find the relationship between the support set and the estimated solution to obtain the objective function model.Finally,combined with the greedy algorithm and the adaptive activation particle mechanism is proposed for location update.The experimental results show that,compared with traditional algorithms,the proposed method more accurately and less errors on the premise of ensuring the integrity of mechanical vibration signals.