Mohammad R. Hoseini

The use of ensemble empirical mode decomposition to improve bispectral analysis for fault detection in rotating machinery

Abstract Empirical mode decomposition (EMD) has been widely applied to analyse signals for the detection of faults in rotating machinery. However, sometimes, it cannot reveal signal characteristics accurately because of the mode mixing problem. Ensemble empirical mode decomposition (EEMD) was developed recently to alleviate the mode mixing problem of EMD. With EEMD, components that are physically meaningful can be extracted from the signals. Bispectrum, a third-order statistic, helps identify phase coupling effects, which are useful for detecting faults in rotating machinery. Utilizing the advantages of EEMD and bispectrum, this article proposes a joint method for detecting such faults. First, original vibration signals collected from rotating machinery are decomposed by EEMD and a set of intrinsic mode functions (IMFs) is produced. Then, the IMFs are reconstructed into new signals using the weighted reconstruction algorithm developed in this article. Finally, the reconstructed signals are analysed via bispectrum to detect faults. The simulation experiments and the physical experiments of two gears with a chipped tooth and a cracked tooth, respectively, demonstrate that the proposed method can detect faults more clearly than can directly performing bispectrum on the original vibration signals.

Understanding vibration properties of a planetary gear set for fault detection

This paper investigates the vibration properties of a planetary gear set. A two-dimensional lumped mass model is developed to simulate the vibration signals of a planetary gear set in the perfect and crack situations. Through dynamic simulation, the vibration signals of each individual component can be simulated, including the vibration signals of the sun gear, each planet gear, and the ring gear. By incorporating the effect of transmission path, resultant vibration signals of the gearbox at the transducer location are obtained. Results show obvious fault symptoms in the signals of an individual component, such as the sun gear. After going through the transmission path, amplitude modulation is shown in the resultant vibration signals. When there is a crack on a sun gear tooth, a large amount of sidebands appears in the vibration spectrum. The locations of these sidebands are investigated and identified, which are helpful for fault detection.

Gearbox Fault Diagnosis Using Two-Dimensional Wavelet Transform

In this paper, a novel technique for denoising gearbox vibration has been proposed. We first convert the vibration signal into a two-dimensional matrix such that each row of the resulting matrix contains exactly one revolution of the gear. This matrix is subsequently denoised using two-dimensional wavelet thresholding method. We apply our proposed method to an experimental data set to investigate the improvement in denoising performance. The experimental data is generated using a test rig on which different damage levels are simulated. The experimental results show that the impulses in the vibration signal can be detected easily from the denoised signal even for slight localized tooth damage. The proposed method is compared to time synchronous averaging and the combination of the time synchronous averaging and the one-dimensional wavelet denoising. The kurtosis value of the denoised signal is used for comparing the denoising performance of these three methods. The comparison study shows that the proposed method outperforms both competing methods, especially in early stages of the fault.

Wear degree prognostics for slurry pumps using support vector machines

Wear damage on impellers is a main cause of the failure of slurry pumps. Prognostics of wear degree allows one to foresee underlying pump failures and thus implement maintenance actions preventively. In this paper, the prediction of wear degree of impellers in slurry pumps is studied. An experimental system is set up to simulate the real working conditions of slurry pumps, from which condition monitoring data and corresponding degrees of impeller damage are collected. An architecture for online prognostics of wear degree is established and an data processing algorithm based on support vector classification is also developed to ensure effective prognostics.

Rotating Machinery Fault Detection Using EEMD and Bispectrum

Ensemble empirical mode decomposition (EEMD) was developed to alleviate the mode-mixing problem in empirical mode decomposition (EMD). With EEMD, the components with physical meaning can be extracted from the signal. The bispectrum, a third-order statistic, helps identify phase-coupling effects, which are useful for detecting faults in rotating machinery. Combining the advantages of EEMD and bispectrum, this paper proposes a new method for detecting such faults. First, the original vibration signals collected from rotating machinery are decomposed by EEMD and a set of intrinsic mode functions (IMFs) is produced. Then, the IMFs are reconstructed into new signals using the weighted reconstruction algorithm developed in this paper. Finally, the reconstructed signals are analyzed via the bispectrum to detect faults. Both simulation examples and gearbox experiments demonstrate that the proposed method can detect gear faults more clearly than can directly performing bispectrum analysis on the original vibration signals.Copyright