Rolling Element Bearing Diagnostics With Higher Order Cyclic Statistics In Variable Speed Conditions

Rolling element bearing(REB)is one of the most important elements for rotating machinery.To guarantee the safety,reliability and efficiency,the detection of incipient defects of REB is necessary and important.The vibration-based diagnostic technique has been studied more study and considered as an effective way.Most of the methods are usually based on the assumption of constant running speed.However,the wheel-bearing experiences small speed fluctuations because of the external load,rail roughness,and wind speed.Moreover,during the deceleration and acceleration process,the bearings would experience large speed fluctuations.The direct analysis of vibration signals of REB with frequency-based methods(such as envelope spectrum analysis)will lead to spectral smearing and false diagnosis.This paper focuses on the third order cyclic statistics theory to explore the bearing fault feature extraction problem under variable speed.The main contents are as follows:By studying the vibration mechanism and failure modes,vibration signals could be divided into deterministic component,of which the behavior can be described exactly,and random component,of which the behavior cannot be predicted.For the deterministic component,it could be further divided into non-Gaussian deterministic components excited by faults,and non-Gaussian deterministic component being irrelevant to faults,which represents the fundamental frequency and its harmonics of the shaft.The latter is mainly caused by misalignment or imbalance in bearing operations.The random component is divided into Gaussian random component and non-Gaussian random component according to the kurtosis index.To reveal the slight change of non-Gaussian nonlinear components in vibration signals,the non-Gaussian random component could be further divided into symmetric and asymmetric non-Gaussian random component,according to the symmetric characteristic of probability density function.Which is by the skewness.To further find the representation for other random components,the matching pursuit combined with genetic algorithm is introduced.This technique could decompose the signal into a linear expansion of waveforms that belong to an atom dictionary.The method about how to describe the components in vibration signal model has been established.The proposed variable speed vibration signal decomposition model is validated by the simulated data,based on the test rig.In order to effectively extract bearing fault feature frequency components under variable speed state,the third order cyclic statistics method has been introduced.In addition,bispectrum can suppress the Gaussian random components and the symmetric non-Gaussian random components of vibration signals,meanwhile retain the information of asymmetric non-Gaussian random components.This was conducive to reducing the interference of noise components and non-faul random vibration components.At the same time,using the sine extraction algorithm to remove non-Gaussian deterministic component with the irrelevant of fault parts.Consequently,algorithms can be successfully employed in fault feature extraction under the complex invalid condition of the REB.In addition,the algorithms proposed in this paper is also competent to the wide range of fluctuation in rolling element bearings comparing with the slice spectral correlation density(SSCD)analysis for weak fault features extraction.The techniques are verified using the experiments on a minor and random slip model but no just suit for minor slip only.