A Method Of Rolling Bearing Fault Diagnosis Based On LCD And VPMCD

Rolling bearing is an element of mechanical equipment which is used most andeasily damaged. When a rolling bearing is broken down, it can cause the failure of thewhole mechanical equipment. So it is very important to monitor and diagnosis theworking conditions of rolling bearing. The key of rolling bearing fault diagnosis isfault feature extraction. When a rolling bearing becomes defective, its signal often hasnon-stationary vibration characteristics. LCD (local characteristic scale decomposition)is an adaptive processing method for non-stationary signal; it can adaptivelydecompose the complex non-stationary signal to the sum of several single componentsignals whose instantaneous frequency has physical meaning. As a result, the LCD isvery suitable for rolling bearing fault signal analysis.Besides，the rolling bearing fault diagnosis is essentially a pattern recognitionprocess. In the field of pattern recognition, VPMCD (variable predictive model basedclass discriminate) is a new method. The method can first establish variable predictionmodels according to the inner relationship between the feature parameters, and then toclassify and recognize the testing data through variable prediction models.In this paper, the LCD and VPMCD are combined, and an approach of rollingbearing based on LCD and VPMCD is proposed. The main research contents of thispaper are as follows:1. The common failure forms of rolling bearings are analyzed and their faultmechanism and vibration characteristics are discussed. Meanwhile, some rollingbearing fault diagnosis methods from two aspects which are signal processing methodsand fault pattern recognition methods are introduced.2. The basic theory of LCD algorithm is studied, and LCD is compared with ITD(intrinsic time-scale decomposition) and EMD (empirical mode decomposition)through simulation signal to prove its advantages. Meanwhile, via experiment, theeffectiveness of LCD in the rolling bearing fault diagnosis is verified.3. Firstly, the basic theory of VPMCD algorithm is studied. Secondly, it iscombined with LCD algorithm and then used in rolling bearing fault diagnosis. Finally,the VPMCD is compared with the neural network algorithm and support vectormachine algorithm, both of which are widely used in the rolling bearing fault patternrecognition, the results show that the VPMCD has the advantage in classification accuracy and training speed.4. According to the drawback in the parameter estimation process of the VPMCD,the VPMCD approach is improved, its parameter estimation approach, least squaremethod, has been replaced by principal component regression. The principalcomponent regression can eliminate the influence of the multiple linear correlationsamong the predictive variables. As a result, it can get more stable model parametersand improve the precision of pattern recognition. The analysis results from the rollingbearing vibration signals of different working conditions demonstrate the effectivenessof the proposed method．5. According to the problem of VPMCD, when its prediction model is established,the model fitting precision can’t significantly enhance with the increase of the capacityof training sample, the original method of polynomial response surface method isimproved. Through the analysis of simulation models, the model fitting precision ofthe improved polynomial response surface method can obviously increase with theincrease of sample size compared with the original method. Therefore, an improvedVPMCD algorithm based on the improved polynomial response surface is proposed bythis paper. Validated by comparing the experimental data, the prediction precision ofthe improved algorithm can rise more obviously with the increase of the trainingsample size.