Mathematical Morphology and Deep Learning-based Approach for Bearing Fault Recognition

Abstract

A fault feature extraction method for rolling element bearings based on mathematical morphology is proposed in this paper. In order to obtain more useful features, this paper attempts to mix mathematical fractal features into time-frequency domain features and wavelet packet energy features. Using the mixed features, support vector machine and deep learning are performed to recognize operation conditions of bearings. It is found that mixed features can improve the conditions recognition accuracy. The comparison results show that deep learning performs better than support the vector machine and is able to predict bearing conditions with a mean accuracy of 99.19%. Therefore, it is concluded that the mixed features and deep learning method are effective for bearing operation conditions recognition.

Fault Diagnosis of Rotating Machinery Based on Multi-Sensor Signals and Median Filter Second-Order Blind Identification (MF-SOBI)

“…where F s = 1/T is the sampling frequency. According to [16], the impulse noise is mainly continuous: (5.36 ± 2.48) × 10 −4 s. For the convenience of calculation and without loss of useful signal, L s = 5.00 × 10 −4 s is assumed. It can be seen from Equation ( 11) that the window width is adaptively adjusted with F s , which is more conducive to eliminating impulse noise and retaining useful signals.…”

“…Therefore, more accurate and comprehensive extraction of vibration signal characteristics has been the pursuit of researchers in this field. The methods of feature extraction of vibration signal include empirical mode decomposition (EMD) [6,7], minimum entropy deconvolution [8,9], an adaptive filter [10,11], matching tracking [12,13], mathematical morphology analysis [14][15][16], cyclostationary signal analysis [17,18], a Wiener filter [19,20], wavelet transform [21][22][23], a Kalman filter [24,25], and stochastic resonance [26,27]-all of which help with the development of mechanical fault diagnosis. The above methods can effectively eliminate background noise and interference components, and extract a fault signal in a specific environment, but are not suitable for complex interference situations.…”

Multisource Fault Signal Separation of Rotating Machinery Based on Wavelet Packet and Fast Independent Component Analysis

“…In order to effectively identify the feature information contained in the fault signal of rotating machinery and reveal its inherent characteristics, many fault feature extraction methods of rotating machinery have been proposed, such as empirical mode decomposition (EMD) [7,8], mathematical morphology filtering [9,10], wavelet decomposition [11,12], adaptive filtering [13,14], matching pursuit [15,16], cyclostationary signal analysis [17,18], Wiener filter [19], Kalman filter [20,21], and stochastic resonance [22,23] that are widely used in early fault diagnosis of rotating machinery. The EMD proposed by Huang et al [7] is a nonstationary signal analysis method, which can find the hidden characteristic information in the signal, and has been widely used in the extraction and noise reduction of the impact signal of rotating machinery.…”

“…Eitner et al [44] used two blind source separation algorithms to estimate the modal parameters of a reduced-scale rocket nozzle using only measurements of deformation. Ge and Jiang [9] proposed a framework based on joint blind source separation (JBSS) in order to solve the jamming suppression problem in the noise environment for the distributed radar with single transmitter and multiple receivers, where the multiple jamming enter into all the receivers through the main beam of the antennas. Belaid et al [45] proposed a new multiscale decomposition algorithm which enables the blind separation of convolutely mixed images.…”

A New Fault Diagnosis Method for Rotating Machinery Based on SCA-FastICA

“…Various traditional modern signal processing methods, such as empirical mode decomposition (EMD) [6], wavelet transform [7,8], adaptive filter [9,10], Kalman filter [11,12], and mathematical morphology analysis [13,14], have been widely used in vibration signal analysis. Qin et al [5] proposed a novel M-band flexible wavelet transform for identifying the underlying fault features in measured signals.…”