The Pattern Recognition Of Mechanical Failure Based On Nonlinear Time Series

Mechanical fault recognition issues includes signal noise reduction, feature extraction, dimension reduction, failure mode of description and identification. Pattern recognition of mechanical failure often analyzes the collected time series. However, the signal of mechanical failure is non-linear and non-stationary, and there is noise. Therefore, it is great significance for finding a pattern recognition method applied to failure of mechanical equipment.The paper combines the phase space reconstruction with the independent component analysis and finds proposed local independent projective noise reduction algorithm. Using this method tests the simulated signal and gets a better noise reduction. It is compared with projective noise reduction algorithm and local projective algorithm. Noise reduction is close to local projective reduction algorithm and better than the global projection. .Low speed and heavy loading bearing vibration signal uses this method and the fault is identified accurately.The three nonlinear dynamics parameters of the correlation dimension, largest Lyapunov entropy index and Kolmogorov set up the feature vector. Using support vector machine as classifier, testing and training the four failure types in the broken gear tooth , wear, pitch error and the normal. The results show that the method has a better recognition.