Planetary gearbox fault feature enhancement based on combined adaptive filter method

Abstract

The reliability of vibration signals acquired from a planetary gear system (the indispensable part of wind turbine gearbox) is directly related to the accuracy of fault diagnosis. The complex operation environment leads to lots of interference signals which are included in the vibration signals. Furthermore, both multiple gears meshing with each other and the differences in transmission rout produce strong nonlinearity in the vibration signals, which makes it difficult to eliminate the noise. This article presents a combined adaptive filter method by taking a delayed signal as reference signal, the SelfAdaptive Noise Cancellation method is adopted to eliminate the white noise. In the meanwhile, by applying Gaussian function to transform the input signal into high-dimension feature-space signal, the kernel least mean square algorithm is used to cancel the nonlinear interference. Effectiveness of the method has been verified by simulation signals and test rig signals. By dealing with simulation signal, the signal-to-noise ratio can be improved around 30 dB (white noise) and the amplitude of nonlinear interference signal can be depressed up to 50%. Experimental results show remarkable improvements and enhance gear fault features.

A Review on Vibration-Based Fault Diagnosis Techniques for Wind Turbine Gearboxes Operating Under Nonstationary Conditions

“…While analyzing the signals using Vold-Kalman filtered order tracking, a longer calculation time is required [105], thus restricting it for real-time processing. Moreover, adaptive filters [106,107], morphological filters [108,109], and various other filters [110][111][112][113] have been researched for gearbox fault diagnosis. These filters have also been used in a cepstral method proposed by Randall and Smith to remove structural response by making the responses insensitive to speed change [114].…”

Gear fault diagnosis using the general linear chirplet transform with vibration and acoustic measurements

“…where Y(x) and W(x) are the respective Fourier transforms of vibration/acoustic signal and Morlet wavelet, and U(x) is the heavy side step function. In this work, the spacing between scales is governed by the octave band analysis or the logarithmically spaced scales as expressed in equations (4) and (5). Here, the number of voices/ octaves of 32 was chosen in this work as it provides a sufficient frequency resolution in the time-frequency representation of vibration and acoustic measurements obtained from the gearbox test-rig…”

“…1,2 Hence, there are limitation of these methods in obtaining accurate information regarding the state-of-health of the gearbox, although they are commonly used for gear health monitoring purposes. 3–8 Thus, signal processing methods with the capability of extracting the non-stationary characteristics of faulty gearbox are expected to be able to improve the accuracy of gear fault diagnosis.…”

Fault diagnosis of driving gear in a battery swapping system based on audio features and SRC-Adaboost

“…For signals with strong noises, previous fault diagnosis methods propose different solutions in the preprocessing stage, feature extraction stage, and pattern recognition stage. In the preprocessing stage, various filters are designed for signal denoising, such as conventional band-pass filters or adaptive filters [18]. In the feature extraction stage, features that are insensitive to noises are extracted, such as symplectic geometry packet decomposition [19], minimum entropy deconvolution [20], and wavelet correlation feature scale entropy [21].…”