| As the science and technology developed,there is an increasing need for rotating machinery in various industries.And rotating units tend to become more automatic,high-speed and upsizing,while the structure is becoming more and more complex.The health status of rolling element bearing(REB)has a highly correspondance to ensure the stable operation of the whole mechanical equipment,for the reason that REB is the key component of rotating machinery.It is of vital importance to monitor the running process of REB to help raising warning and halting machine,for the capacity that it can elude some big safety incidents.In the mean time,carrying out the fault type identification when the fault occurs,and verifying the fault location and fault severity,can make a good instruction to the staff to develop a scientific and reasonable maintenance plan,and further reduce the downtime and production costs.Thus,aimed at REB,a two-stage task of health status assessment and fault identification was separated from the diagnose process.The research on the fault diagnosis technology of REB that combined local mean decomposition and graph modeling method was proposed,and the method was verified by using the existing open source bearing data sets and simulation signals.Firstly,based on the collected vibration signals of REB,the adaptive time-frequency analysis was carried out by local mean decomposition(LMD),thus the information of the original signals from multiple scales can be derived.Considering the shortcomings when directly using LMD to extract low dimensional efficient information and fault characteristics,the undirected weighted graph modeling strategy that based on the graph theory was employed,for which together with the multi-scale component information of the signal to construct dynamic representation model of the raw vibration signals,and mining the implicit fault characteristics during bearing operation process.Through the process of graph modeling,the original vibration signals were transformed from one-dimensional time domain to graph series.In the stage of health status assessment,the graph similarity metric is employed to fuse the multi-scale information embedded in graph,and indeed to extract the dynamic health index of REB.Using the generated dynamic health index,the health status assessment is carried on to implement the fault detection and warning in the early stage of REB.The effectiveness of dynamic health index and decision algorithm has been analyzed and demonstrated through the simulation signal.Once a fault warning has been raised in the stage of health status assessment,the method will be transferred to the fault identification stage.The fault graph model was extracted firstly,then using principle component analysis(PCA)method,the dimension reduction work was carried towards to the graph structure.The compressed data were taken as the characteristics under different fault types,and together with pattern recognition classifier,the fault location and fault size were classified to fulfill the fault identification of REB.The capacity of this work that it maintain the main powerful information while keeping data at a lower dimension has been validated through simulation signals.To sum up,multi-scale analysis of REB vibration signals was carried out by LMD in this paper,on the basis of the knowledge of graph theory,the effective data modeling method was researched in order to achieve accurate and dynamic characterization of raw data.A two-stage method that combined health status assessment and fault identification was proposed to carry out fault diagnosis of REB.The capacity of the method was validated and analyzed by the open source bearing data sets,for which provided by Xi’an Jiaotong University and Case Western Reserve University,respectively.Finally,the conclusion of this study and discussion of the future work was made.This work is supported by Natural Science Foundation of Shandong Province,China(ZR2019MEE063). |
