Research On Fault Diagnosis Method Of Rotating Machinery Based On VPMCD

Rotating machinery is the most common used equipment in industry. Because of thehard working environment, strong vibration and noise, it’s easily damaged. When therotating machinery is broken down, it can cause the failure of the entire rotatingmechanical equipment, even leading the entire production chain to paralysis and causinghuge economic losses. So it’s meaningful and valuable to monitor and diagnosis theworking conditions of rotating machinery.The rotating machinery fault diagnosis is essentially a pattern recognition process.In the field of pattern recognition, variable predictive model based class discriminate(VPMCD) is a new method. The method can first establish variable prediction modelsaccording to the inner relationship between the feature parameters, and then to classifyand recognize the testing data through variable prediction models. In rotating machineryfault diagnosis, there is certain internal variable relationship among characteristicparameters; but the relationship has different characteristics among different systems orcategories. Therefore, VPMCD can be applied in rotating machinery fault patternrecognition.In this paper, the method of rotating machinery fault diagnosis based on VPMCD isproposed. The main research contents of this paper are as follows:1. The basic theory of VPMCD algorithm is studied. The VPMCD algorithm is madea comparison with the neural network and support vector machine. The results of thecomparisons indicate the suitability and effectiveness of the method.2. The applications of the VPMCD algorithm in rotating machinery fault diagnosisare studied. In this paper, sample entropy, permutation entropy and independentcomponent analysis (ICA) are applied as fault feature extraction methods to the faultdiagnosis. Respectively, the VPMCD and sample entropy are used for gear faultdiagnosis, the permutation entropy and VPMCD are used for rolling bearing faultdiagnosis, the correlation coefficient of independent component analysis and VPMCDare used for rolling bearing fault diagnosis. The analysis results from the rotatingmachinery vibration signals of different working conditions and fault types demonstratethat the VPMCD can be effectively applied in rotating machinery fault patternrecognition.3. VPMCD algorithm classifies the faults by the minimum prediction error sum of squares value. So it not only can be applied to a variety of samples of pattern recognition,but also can be applied to the pattern recognition of single samples. Based on that, thepaper proposed a new one-class classification method—One-class variable predictivemodel based class discriminate (OC-VPMCD). It classifies the test samples bycomparing the prediction error sum of squares value with the prediction error sum ofsquares threshold value. Validated by the experimental data, the result shows thatOC-VPMCD can be effectively applied to the pattern recognition of single samples.