Session: 05-02 Topics on Gas Turbine Diagnostics

Paper Number: 83600

83600 - Bearing Diagnostics Using Kurtosis Spectral Correlation Based on Cyclic Modulation Spectrum Estimation 

Condition-based maintenance is a key strategy in industry for fault detection and diagnosis to increase cost effectiveness of rotating machinery. Early detection of bearings faults is fundamental to avoid unexpected failures in heavy and complex rotating machinery, such as helicopter gearboxes or wind turbines.

 

Vibration signals are widely acquired from these structures in order to estimate the health condition of bearings, as microphone signals acquire higher levels of background noise and acoustic emission sensors are less cost effective. Bearing damage signatures are, however, usually masked by other dominant sources, such as gears meshing. Thus signal processing methods are required to extract and detect the faulty signature of bearings present in the signals.

 

Classic methods are based on the envelope analysis, where the resonant frequency excited by the fault impulse is selected as a carrier, and a band pass filter is applied around the carrier. Then, the spectrum of the envelope of the filtered signal (e.g. via Hilbert transform), results in the demodulated spectrum (a.k.a. Squared Envelope Spectrum (SES)) with peak detection at the low frequency corresponding to the characteristic frequencies of the bearing related to the damage. Such frequencies can be the Ball Pass Frequency of the Outer Race (BPFO) or Inner Race (BPFI), and when the presence of these frequencies is detected on the spectra, diagnosis of faulty bearing is performed.

 

Carrier frequency bands can be selected manually based on engineering knowledge, or by automated band selection tools. As the damage of the bearings leads to cyclic impulses by nature, the Kurtosis of the signal tends to increase in the presence of localised surface damage. Following this reasoning,  Spectral Kurtosis was initially proposed for selection of the carrier, where the kurtosis level at different frequency is extracted and the highest value corresponds to the optimal band for filtering. The method was extended to the Fast Kurtogram, as a more computationally efficient procedure. Several band selection tools based on the extraction of the impulsive nature of the bearing damage via Kurtosis, Entropy and other sparsity measures have been further proposed. On the other hand, these methods lead often to the selection of unrelated carriers when other cyclostationary impulses (e.g. due to Electromagnetic Interference) or non-cyclostationary impulses are present.

 

More recently, cyclostationary tools like the Spectral Correlation have been proposed due to their high performance in extracting the hidden signatures of bearing damages. This method describes the signals in the Frequency-Frequency domain, where the carrier frequencies in the spectral frequency axis are correlated to their cyclic modulation frequency. The Spectral Coherence is a normalization applied on the Spectral Correlation by the Power Spectrum (at zero cyclic frequency), resulting in enhanced peak detection at higher frequency bandwidth, which are commonly bearing fault related.

 

This paper proposes a novel cyclostationary tool in the Frequency-Frequency domain, termed as the Kurtosis Spectral Correlation. The method is based on the autocorrelation of the Spectral Kurtosis with a time-lag, resulting in the Kurtosis level in the Frequency-Frequency domain via the Cyclic Modulation Spectrum estimation. The method is proposed as a method specific to detect cyclic impulsive signatures (related to bearings), and reduce 1^st order cyclostationary (related to gears) and non-cyclostationary impulses. The method is tested and validated on real signals captured on an experimental drivetrain and complex structures, such as wind turbines and helicopters.

Presenting Author: Alexandre Mauricio LMSD - Mecha(tro)nic System Dynamics, Department of Mechanical Engineering, KU Leuven, Belgium

Presenting Author Biography: ...

Authors:

Alexandre Mauricio LMSD - Mecha(tro)nic System Dynamics, Department of Mechanical Engineering, KU Leuven, Belgium
Konstantinos Gryllias LMSD - Mecha(tro)nic System Dynamics, Department of Mechanical Engineering, KU Leuven, Belgium