Research On The Application Of Blind Source Separation In Fault Diagnosis Of Gear Box

As the most common mechanical equipment, gear box is applied widely in industry,agriculture, transportation, petroleum and other industries. The vibration signal of gear boxshows the information of its running state, which will change as the running state of gear boxchanges. A research on the vibratory characteristics of gear box and the analyses of theeffective vibration signals can be a great assistance of condition monitoring, fault diagnosisand trouble shooting, which can strengthen the safety and reliability of a working gear box,reduce its maintenance costs and avoid unnecessary lose. It is of vital importance inpreventing unexpected accidents and maintaining the normal operation. Blind sourceseparation (BSS) is a new method of signal processing in recent years. In contrast totraditional ways, BSS can extract or recover source signal from observed mixed signalwithout any useful information of source signal and transmission path. This thesis is about thesimulation of the common fault of gear box in the condition of laboratory, disposing the faultsignals of gear box by blind source separation and intelligently identifying the faultycondition of gear box by the method of support vector machine (SUV) after extractingeigenvector, which achieves success.Fault diagnosis of gear box mainly includes diagnostic information acquisition,fault information eigenvalue extraction and fault intelligent identification, which takes theextraction of fault information eigenvalue and fault intelligent identification as the main part.Although the vibration signal will change in the occurrence of fault, the characteristic offaulty signal is very weak in contrast with that of vibration signal and normal meshing signalof gear box, which is difficult to reflect on the curve graph of signal time-frequency domain.The blind source separation mentioned in this thesis can separate the testing signal which isunknown source signal from the fault source signal of the broken-down gear box and estimate the fault type of the gear box based on the theoretical calculation.Support vector machine is a new method of machine learning, which is based onstatistical learning theory and structural risk minimization. This kind of method ischaracterized high degree of theoretical rigor and strong adaptability, which is applied inpattern recognition widely. In the end of this thesis, the testers get a fault recognition ratewhich reaches up to more than85%by adopting the support vector machine to identify thefault of gear box intelligently, which presents the feasibility of the application of SVM andintelligent identification of the fault of gear box.