Classification Of Weak Rolling Bearing Faults Based On Morphological Filtering And Local Tangent Space Alignment

Rolling bearing is one of the most widely used parts in the rotatingmachinery, which plays an important role, but also is very prone to malfunction.The health of rolling bearing affects the working condition of the entiremechanical system. The presence of rolling bearing failure will not only affectthe normal use of the device, but also makes the normal production process isdifficult to achieve. If the fault is not found in time, severe cases can lead toaccidents and significant adverse effects on the people’s property and personal.Therefore, the study of fault diagnosis of rolling bearing has a very importantsignificance.Vibration is an inevitable phenomenon bearing runtime. The vibration ofthe rolling bearing tends to be more intense when a fault emerges from therolling bearing’s outer ring, inner ring or rolling elements. The periodic pulsesignal will be accompanied by the vibration which results in production ofmodulation signal. Different fault presents different frequency. So one of the keyin the current research in this field is how to acquire effectively the fault signal characteristics.Due to the influence of the random noise, vibration transmission path andother factors, the vibration signal of bearings in actual work process is nonlinear.And it leads to the traditional linear methods such as Fourier transform, wavelettransform is difficult to accurately extract the features of the fault signal. Thispaper briefly introduces several kinds of time domain, frequency domain andtime-frequency domain methods and expounds the mathematical morphologymethod in dealing with the application of the nonlinear vibration signal and itsadvantages and disadvantages. At the same time, in view of the weak faultoccurs, rolling bearing vibration signals which have high noise, lowsignal-to-noise ratio, this paper further illustrates the manifold learning how tobe applied in the case of fault classification. It also compares the global andlocal manifold learning method.In this paper, the theory is verified by experimental data. The experimentsrespectively collected from various parts of the rolling bearing’s inner race,outer race and rolling elements. The diameters of the fault are0.1778mm,0.3556mm,0.5334mm and0.7112mm. The fault types of pitting. And then withthe help of MATLAB software programming, to deal with the data fromexperiment.Based on the experimental data, respectively, using traditional waveletpacket filtering and morphological filter to filter nonlinear signal noise data, andthen further using of global and local manifold learning algorithms for data extraction after filtering noise eigenvalues and fault classification. The effectiveprossing results illustrate the advantages of the morphological filtering and localtangent space alignment algorithm in dealing with nonlinear signal.