Mechanical Siganl Processing Based On Sparse Representation And Its Application In The Diagnosis Of Rolling-element Bearings

With the development of society and the continuous improvement of productivity,modern industrial equipment tends to have the property such as integration,automation and high-speed.However,the increased functionality of the device means that the com-plexity of the device is also increasing,which leads to even a small failure is likely to cause incalculable consequences.In order to ensure the operation of equipment and re-duce the impact of its accident on people's life,fault diagnosis technology is extremely important.Accurate identification and prediction of mechanical failure can not only protect the personal safety of life and property,but also reduce the cost of enterprises in the maintenance,which brings higher economic efficiency.In addition,the develop-ment of fault diagnosis technology makes the production of toxic,harmful,flammable,explosive and other dangerous products more secure,thus maintaining the stability of region and even country.In all mechanical failures,45%to 55%of the total is caused by rolling bearings.Bearing wear,spalling,pitting,cracks and other failures will lead to system crashes,thereby reducing the reliability of the system or even disastrous con-sequences.Therefore,the implementation of effective diagnostic strategies to reduce the bearing failure caused by the loss has become a top priority.This study designs a series of new methods for bearing signal processing and fault recognition based on sparse representation theory.In the research,this paper will com-bine the latest advances in sparse representation with bearing fault diagnosis to improve the traditional diagnostic strategies,which aims at acquiring higher diagnostic accu-racy.In order to achieve this goal,the article will mainly start from two aspects.On the one hand,the theory of sparse representation is discussed in depth,grasping the latest developments in the field and getting to know the sparse optimization methods at the forefront.On the other hand,the fault generation mechanism of rolling bearing sig-nal is researched,and a sparse method suitable for fault feature extraction is designed.For each new method,the article will make a comparition with traditional methods us-ing simulation and experiments to verify the validity and superiority of the proposed algorithm.The main contents of this paper include:1.The sparse representation theory is studied comprehensively,including the the-oretical basis,sparse optimization algorithm and the dictionary learning method.In this process,the applicable conditions and advantages of various algorithms are analyzed,so as to design a sparse optimization method suitable for mechanical signal processing and bearing feature extraction.2.The application of sparse representation with wavelet basis in bearing fault diag-nosis is studied using overcomplete wavelet transform.In this paper,two over-complete wavelet bases are respectively selected from time-domain redundancy and frequency-domain redundancy,and the fault signal is sparsely coded and re-constructed by the ADMM method.Simulation and experimental results show that,compared to the traditional wavelet threshold denoising,the sparsely op-timized and reconstructed signal has less noise,which makes the fault feature more prominent,thus creating a more favorable condition for fault type recog-nition.This result confirms the effectiveness of sparse representation in bearing fault diagnosis.3.The dictionary is trained by KSVD to study the effect of sparse representation based on learning dictionary in bearing health state recognition.Compared with the fixed base,the learning dictionary can detect more accurate data structure and provide more sparse coding,which is conducive to fault feature extraction.Based on the KSVD method,a complete training dictionary based fault diagnosis procedure is designed,and the parameters selection in the practical application is analyzed in theory and experiment.Through the simulation and diagnosis ex-periments,this paper verifies that the KSVD dictionary can accurately detect the impact characteristic of the bearing fault signal and obtain a better diagnosis result than the sparse representation of the fixed base.4.Based on the group characteristics of bearing fault signals,the fault diagnosis method of rolling bearing based on group sparse is studied.The impact of rolling bearing failure are not independent,but will form a cluster within a certain range of shock.Aiming at this situation,this paper proposes an impulse-detect sparse coding algorithm.The algorithm achieves more accurate coding by considering such group information in the sparse optimization process,thus detecting the fault information hidden in the noise.Simulation and experimental results show that the sparse detection algorithm has high precision and high efficiency.Even if it is difficult to find accurate sparse parameters,the high efficiency of the algorithm allows it to be performed several times to find the most suitable parameter,thus achieving a satisfying diagnostic result.5.Based on the periodic characteristics of bearing fault signals,the diagnosis method using low rank optimization is studied.The periodicity of the fault signal of the rolling bearing means that the signal will repeat to some extent at every interval.If a matrix is composed of these similar signal segments,the matrix will ideally have a low rank property.Inspired by this idea,this paper presents a low-rank impulse detection algorithm.The algorithm uses the clustering and nuclear norm to optimize the fault signal,and finally highlights the fault characteristics,and realizes the fault diagnosis.In this paper,a comprehensive evaluation of the al-gorithm is carried out by simulation and experiment,which show the excellent noise reduction ability and high precision diagnosis effect of the algorithm.