New Approach of Wavelet Fractal Analysis to Mechanical Fault Diagnosis

Abstract

Wavelet transform decomposes signals in both time and frequency domains. According to the principle of multi-resolution, irregularities or complexities of decomposed signals in various frequency bands are different, which can indicate mechanical fault. Fractal dimension of decomposed signal is an excellent factor to express the irregularity or complexity. It can make features of mechanical faults remarkable. Based on wavelet fractal dimension, successful diagnosis of looseness fault of bearing bridge for 50MW turbo-generator was obtained.

Fault diagnosis based on optimized node entropy using lifting wavelet packet transform and genetic algorithms

“…Another point of interest is that the feature sets extracted from the third level of the WPT decomposition tree (entropies of the third level, WPT-Energy) performed better than others when using NNs as classifiers. This may explain why three-level WPT is very popular in the literature [4][5][6][7][8][9][10][11].…”

“…As such, exactly depicting the irregularity and complexity of the measured vibration signals is crucial for successful implementation of fault diagnosis. Considering the similarity in the principle of the multi-resolution analysis in WPT and fractal analysis, the fractal dimension (FrD) is taken as a measure to quantify the irregularity and complexity of the node signal components resulting from WPT in references [9] to [11]. The combined use of WPT and the fractal analysis (WPT-FrD) can accommodate not only the non-stationarity but also the irregularity and com-Fig.…”

“…The combined use of WPT and the fractal analysis (WPT-FrD) can accommodate not only the non-stationarity but also the irregularity and com-Fig. 1 An example of the decomposition tree of a three-level WPT, where nodes with diagonal background shading are those that WT is able to obtain plexity inherent in the vibration signals, which has been evidenced by their successful applications in the diagnosis of a real turbogenerator and a laboratory rotor test rig [9][10][11].…”

Comparisons of different deep learning-based methods on fault diagnosis for geared system

“…The linear stationary mathematical transformation methods are represented by time domain statistical analysis, 1 frequency domain statistical analysis, 2 Fourier transform analysis, 3 time series model analysis method, 4 refined spectrum analysis, 5 holographic spectrum analysis, 6 singular spectrum noise reduction method, 7 matching tracking analysis, 8 and geometric fractal analysis method. 9 Non-stationary, non-Gaussian distribution, and nonlinear random signal processing methods are represented by high-order spectral analysis, 10 principal component analysis, 11 short-time Fourier transform (STFT), 12 Wigner–Ville distributing, 13 wavelet transform (WT), 14 cyclic stationary analysis method, 15 random resonance method, 16 empirical mode decomposition (EMD), 17 Hilbert–Huang transform (HHT), 18 and the second-generation WT method. 19 Intelligent diagnosis technology methods mainly include intelligent diagnosis technology based on an expert system, 20 intelligent diagnosis technology based on an artificial neural network, 21 intelligent diagnosis technology based on a fuzzy logic, 22 intelligent diagnosis technology based on a genetic algorithm, 23 fault diagnosis technology based on a fuzzy neural network, 24 and so on.…”