Research On Mechanical Fault Response Characterization Based On Sparse Decomposition And Its Fast Mining Method

Mechanical fault response mining refers to the process to find out the positions of multiple responses of the same sources in a time series according to the pattern similarity and distribution characteristics of the multiple responses aroused by the same sources.This is the primary task to realize the separation of multiple vibration sources in condition monitoring.In order to overcome the problem of mode aliasing and noise disturbance,we need to do research on the representation of time series fragments.,it is also necessary to research on the similarity measurement method of spatial point group to find out measurement methods corresponding to mechanical fault response characterization.In demand of large amount of monitoring data and real-time processing,we should use data mining to find a fast mining method.The overall research of this paper is mainly carried out from the three aspects below:1.Many mechanical devices need to run for a long time.In order to ensure the operational reliability and safety of the key electromechanical systems,it is critical to perceive the performance effectiveness of the device.Vibration time series carries the internal condition information of the machines and devices.Being expressed only in time series is too rough for the mechanical fault diagnosis model to distinguish similar fault mode and judge the subtle changes in degree of damage.In order to reveal the multi-dimensional information in fault responses in all aspects so that we could obtain more homologous response for distribution reasoning in the test stage,this paper presents a sparse-decomposition-based MP mechanical fault mode characterization method,characterizing the mechanical fault mode into 30 groups of five dimensional atoms which includes the amplitude,frequency,phase and displacement.This method is visualized in form of spatial point cluster scatter diagram matrix,and it forms 10 different two-dimensional spatial point group representation schemes.It is verified by the measured signal that the classification effect of the representation scheme formed by the combination of sparse coefficient and displacement dimension has the best pattern classification effect among the 10 schemes,and there is a stable threshold interval.Compared with the previous characterization,it can use the essential physical characteristics of vibration signal to dig out more homologous response fragments more accurately.2.Aiming at the mechanical fault response represented in the form of spatial point group after MP sparse decomposition,In order to quantitatively analyze the advantages of this characterization method,the similarity relations among the fault responses are quantitatively analyzed using measurement algorithms of spatial point group target similarity relation in the field of geographical science.This paper completes the compilation of five similarity calculation algorithms and draws the conclusion that the distanee similarity has the greatest influence on the realization of fault response pattern classification of rolling bearing through analysis.The study on this algorithm enriches the similarity measurement method of fault response and opens up a new way for the similarity measurement of mechanical fault diagnosis.3.This paper proposes a fast mining method of homologous responses.Firstly,based on the Matrix Profile algorithm,the preliminary mining of homologous response fragments is realized on the time series.Secondly,a homologous response distribution constraint condition is proposed.Based on a small number of homologous response fragments preliminarily mined by MP algorithm,more than 80%homologous response fragments in time series were obtained through calculation and by reasoning from position information,so as to improve algorithm efficiency and realize fault diagnosis more reliably.Finally,the data of the faulty bearings is used to verify the effectiveness of the homologous response fast mining method.The computational performance of the algorithm is verified from three aspects.The result shows that the algorithm could quickly and efficiently mine the homologous response fragnents in the time series.Generally,this paper presents a mechanical fault characterization method based on MP sparse decomposition focusing on the subject of sub-source monitoring,and provides a spatial point group target similarity measurement for the characterization results to do quantitative analysis.The paper finished the mode detection stage of the fault response sub-source monitoring,which provides a new perspective of the study and understanding of mechanical fault.In this paper,a fast mining algorithm for homologous response is proposed based on the Matrix Profile algorithm.Along with the homologous response constraints proposed in this paper,it forms a complete fault response sub-source monitoring study on time series.