A Method Of Roller Bearing Fault Diagnosis Based On Local Characteristic-scale Decomposition

Roller bearing is used very extensively in machinery and equipment and it iseasily damaged. When roller bearing is broken down, it will cause the failure of theequipment, and even lead to an accident. So it has an essential significance to monitorthe working status and diagnose the fault. The key of roller bearing fault diagnosis isthe fault feature extraction. Roller bearing’s fault often leads its vibration signalexhibiting non-stationary characteristics. LCD (Local characteristic-scaledecomposition) is an adaptive processing method for non-stationary signal; it canadaptively decompose the complex non-stationary signal to a series of singlecomponent signals whose instantaneous frequency has physical meaning. As a result,the method is very suitable for roller bearing fault signal analysis and processing.On this basis, the LCD is introduced into roller bearing fault diagnosis, and alsocombined with SVM(Support vector machine), VPMCD(Variable predictive modelbased class discriminate). In this paper, an approach of roller bearing fault diagnosisis proposed. Its main research contents are as follows:1. The common failure forms of roller bearing are introduced and its faultmechanism are discussed. Meanwhile, the basic theory of LCD method is alsointroduced, and LCD(Local characteristic-scale decomposition) is compared withITD(Intrinsic time-scale decomposition) and EMD(Empirical mode decomposition)through simulation signal to prove its advantage in reducing signal distortion anditeration number. And the effectiveness of LCD in processing roller bearing vibrationsignal via experiment.2. Based on the definition of local Hilbert marginal energy spectrum, a faultfeature extraction method for roller bearings is further proposed based on LCD andlocal Hilbert marginal energy spectrum. By using LCD, an original roller bearingvibration signal could be adaptively decomposed into a number of ISCs (Intrinsicscale components), and then the time-frequency distribution could be obtained byapplying Hilbert demodulation to all the components. Then the signal energy over thisfrequency band could be computed subsequently and regarded as the fault featureparameter. Then the feature vector composed by fault feature parameter is input intoSVM, the analysis results show that the proposed approach can effectively extract thefault feature information. 3. Aiming at the noise problem of roller bearing vibration signal, a noisereduction method based on LCD_GRA is proposed, and VPMCD recognition methodis combined to be used into roller bearing fault diagnosis. VPMCD is a new patternrecognition method, and its superiority has been demonstrated by UCI standard data.Through the analysis of simulation and experiment, the effectiveness of LCD_GRAnoise reduction method is verified, and it can be used for roller bearing faultdiagnosis.