Instantaneous Amplitude-Frequency Feature Extraction for Rotor Fault Based on BEMD and Hilbert Transform

Huang, Chuanjin; Song, Haijun; Lei, Wenping; Niu, Zhanya; Meng, Yajun

doi:https://doi.org/10.1155/2019/1639139

Shock and Vibration

On this page

Abstract Introduction Discussion Conclusions Data Availability Conflicts of Interest Authors’ Contributions Acknowledgments References Copyright Related Articles

Research Article | Open Access

Volume 2019 | Article ID 1639139 | https://doi.org/10.1155/2019/1639139

Instantaneous Amplitude-Frequency Feature Extraction for Rotor Fault Based on BEMD and Hilbert Transform

Chuanjin Huang,¹Haijun Song,¹Wenping Lei,²Zhanya Niu,³and Yajun Meng¹

Academic Editor: Davood Younesian

Received26 Oct 2018

Revised09 Jan 2019

Accepted19 Feb 2019

Published12 Mar 2019

Abstract

The vibration signals propagating in different directions from rotating machines can contain a variety of characteristic information. A novel feature extraction method based on bivariate empirical mode decomposition (BEMD) for rotor is proposed to comprehensively extract the fault features. In this work, the number of signal projection directions is determined through simulation, and the energy end condition based on the energy threshold is increased using BEMD to enhance the decomposition quality. Mixed vibration signals are generated along two orthogonal directions. Then, the acquired vibration signal can be decomposed into several intrinsic mode functions (IMFs) at the rotational speed using the BEMD method. Furthermore, the instantaneous frequency and instantaneous amplitude of the real signals and the imaginary part of the IMF signals are obtained using the Hilbert transform. The fault features along two and three dimensions can be investigated, providing more comprehensive information to aid in the fault diagnosis of rotor. Experimental results on oil film oscillation, the oil whirl, the bistability of the rotor, and looseness and rotor rubbing composite fault indicate the effectiveness of the proposed method.

1. Introduction

The features of vibration signals form the basis of fault diagnosis of rotating machinery. A variety of common fault feature extraction methods, such as the wavelet transform (WT) [1], empirical mode decomposition (EMD) [2], and local mean decomposition (LMD) [3, 4], are widely used for such diagnosis. In the WT, the wavelet function must be preset, which leads to limitations in the self-adaptability of the WT. The EMD and LMD methods adaptively decompose the signals into a series of intrinsic mode functions (IMFs) and product functions (PFs), respectively, according to the characteristics of the signal envelope. However, these traditional methods are suitable only for processing real-value data on a single channel [5].

The fault vibration characteristics from two orthogonal sensors, which are placed in various locations on the machine, have been shown to be important for diagnosing faults. To improve the reliability of the diagnostic results, the feature extraction method based on multichannel signal features is becoming increasingly more popular.

The orbit of each harmonic from the vibration signals is an ellipse when the rotor runs in a steady state. The fault feature extraction methods based on homology information, such as the holospectrum [6], full spectrum [7], and full vector spectrum [8], have been proposed in previous reports. The vibration signals in the orthogonal direction can form an ellipse. Because the homologous information incorporates the orthogonal direction of the vibration signal features, the diagnostic results are more comprehensive and accurate [9]. Fault features can be described by the homologous information through round or elliptical information. Unfortunately, in most cases, the circle or ellipse information can be extracted using only a Fourier transform. Therefore, the homologous information method is applicable primarily to the analysis of stationary signals.

The EMD or LMD methods can locally smooth the nonlinear signal. Homologous information technology combined with EMD and LMD [10, 11] has been proposed to extract the features of the nonlinear signal. However, the EMD and LMD methods decompose the signals from different directional sensors individually and separately. The circle or ellipse information is calculated according to the one-to-one mode principle [11]. When the EMD or LMD is used to decompose the orthogonal vibration signals, the number of decomposition result is nonuniform, which causes the information fusion to become increasingly challenging [12]. In addition, the EMD or LMD method is relatively sensitive to noise, which causes the IMF group or PF group frequency to be inconsistent with other IMF or PF groups.

To improve the capacity of EMD and LMD, these methods have been extended recently to process multivariate signals. Some examples include bivariate empirical mode decomposition (BEMD) [13], complex local mean decomposition (CLMD) [14], trivariate EMD (TEMD) [15], and multivariate EMD (MEMD) [16]. However, these extended methods were not specifically developed for machine fault feature extraction. In BEMD, the local mean of the bivariate signal is calculated by projecting the signal to a number of directions. Following the same idea, the TEMD and MEMD were proposed, respectively, for three-dimensional and n-dimensional signals. The CLMD estimates the local mean by projecting only a bivariate signal onto the x- and y-axes. The bivariate EMD has been employed to detect wind turbine mechanical and electrical faults by decomposing the fault signals [5]. Here, the electrical signal was analyzed without considering the noise and the number of BEMD projections. The associated experiments indicated that the number of projection directions affects the decomposition results, particularly when noise is included. In addition, additional false components were identified in [5]. The MEMD has also been employed to fault diagnosis of rolling bearing [17]. Methods based on BEMD or MEMD are used to analyze and extract signal characteristics in multichannels. However, the joint information between multiple sensors was not considered in [5, 17]. This method, combined with the CLMD or MEMD and full spectrum, has been proposed to obtain joint information among multichannel signals [12, 18]. The methods based on CLMD or BEMD simultaneously decompose the signals into multichannels, thus ensuring that the number of decomposition results is the same and is readily incorporated. However, the instantaneous feature cannot be extracted due to the elliptical information that was generated using the conventional Fourier transform.

The signals are analyzed as the superposition of slow and fast oscillations in the EMD, and the bivariate extension and bivariate signals, e.g., the orthogonal vibrations, are analyzed as the superposition of IMFs at the rotational speed in the BEMD. To extract the instantaneous vibration characteristics and the joint information among the multichannel signals, the fault feature extraction method based on the BEMD is proposed to decompose the multicomponent rotation signal into a series of single-component rotation signals. The instantaneous amplitude and instantaneous frequency of the IMFs are further obtained using the Hilbert transform (HT).

2. Brief Description and Improvement of BEMD

2.1. Brief Description of BEMD

The fundamental concept of BEMD is that bivariate signals are made up of slow rotation signals and fast rotation signals superimposed on the slow rotation signals [13]. For a mixed signal , the decomposition process based on the BEMD is as follows [5]: Step 1: determine the number of projections N and calculate the projection directions: Step 2: project the complex-valued signal on the directions : Step 3: extract all local maxima of . Herein, indicates the number order of individual local maxima. Step 4: interpolate the set by spline interpolation to obtain the tangent along the direction . Step 5: repeat 2–4 until the tangents in all N projections are obtained. Step 6: compute the mean of all tangents: Step 7: subtract from to obtain , i.e., Step 8: perform sifting process whether the stopping criterion similar to the one proposed in [19] is met; however, and are bivariate signals. If not, regard as original signal and repeat Steps 2–7 until the stopping criterion is met. Step 9: record the obtained IMF and remove it from , i.e., Step 10: take as the original signal and repeat the above calculation until the second IMF is obtained. The remainder is then calculated as follows: Step 11: iterate the previous calculations until acquiring all IMFs contained in . After the above process, can be expressed as follows:where represents the total number of IMFs.

During the decomposition process, the bivariate rotating signal is required to rotate around the zero point. The rotating machinery often revolves around a central movement, which coincides with the requirements of the BEMD. The present study takes the typical Jeffcott rigid rotor-bearing system dynamics equation to structure the complex z (z = x + jy) with x and y representing the vibration signals collected by two orthogonal sensors. The obtained characteristics curve is shown in Figure 1. The BEMD analysis of the with the parameter N of 4 is shown in Figure 2, which shows that the target signal is decomposed into bivariate rotation components at the rotation speed, allowing investigation of the instantaneous amplitude-frequency (IAF) characteristics of the main components.

(a)

(b)

(a)

(b)

(c)

2.2. Improved BEMD Method

In the original BEMD algorithm, the loop cutoff condition is such that all IMF components are obtained from the original signal, which leads to additional IMF components and increases the computing time of BEMD. In most cases, the fault characteristics are primarily contained in the IMFs with a higher energy, and the vibration faults corresponding to the rotational modes are limited. In this study, the end condition of the BEMD based on the energy threshold is proposed based on the reasons mentioned above. A ratio λ is set, and the ratio between the signal to be decomposed and the energy of the original signal is less than a specific value that serves as a criterion to stop the BEMD algorithm. λ is calculated using the following formula:

3. IAF Feature Extraction of the Bivariate Rotation Signal

As noted in Section 2, BEMD decomposes the complex-valued signal of the multiple components into the complex-valued signals of a single component. is a complex-valued signal with . represents the horizontal component of the vibration signal, and represents the vertical component of the vibration signal. The HT is a classical method to obtain the IAF of the signal. The real and the imaginary components of are transformed with the HT to obtain the IAF. The following formulae are established:

The corresponding resolution signal expressions arewhere and are the instantaneous amplitudes of the analytic signals of and , respectively, and are the phase functions of and , respectively, and

The instantaneous frequency function is derived from the phase function:

We define and as the instantaneous amplitude and the instantaneous frequency of , respectively, and define and as the instantaneous amplitude and the instantaneous frequency of , respectively.

4. Experiment Verification

4.1. Analysis of the Oil Film Oscillation Signal

The rotor oil film oscillation test devices are shown in Figure 3. The signals are collected using eddy current sensors along the orthogonal direction to the axis. The first-order critical speed is found to be in the range of 3,200 to 3,400 rpm using the resonance phenomena by increasing the speed of the rotor test rig whose speed is increased from 0 to 8,331 rpm and then decreased to 0. Certain oil film vortex and oil film oscillation phenomena occur during the experiment. The data recording device uses the INV306 collector, which has a 16-bit 4-way parallel AD converter. The sampling frequency of the collector was set to 2,048 Hz in the following experiments.

The oil film oscillation signals along x and y (where the rotor speed is 6,480 rpm) in the horizontal and vertical directions are collected by eddy current sensors. x and y and their Fourier spectra are shown in Figure 4, where X = 108 Hz is the frequency of the fundamental frequency signal. Let z = x + jy. The three-dimensional time-domain waveform and the two-dimensional plots of z are shown in Figure 5. Figures 4(b) and 4(d) indicate that the amplitude of the 0.48X component is more prominent, particularly in the vertical direction and even exceeds the amplitude of the fundamental frequency signal. Figure 5 indicates that the amplitude of the oscillation signals is variable but that the Fourier spectra cannot distinguish this change.

(a)

(b)

(c)

(d)

(a)

(b)

The EMD is applied to decompose the signals x and y, as shown in Figure 6. The S1 and S2 columns represent the decomposition results of x and y, respectively. The number of IMFs generated from signal x is 10, but there are only 8 from signal y. The mismatch in the number of IMFs leads to challenges in information fusion. In addition, modal aliasing occurs in IMF₁ from signal x. The mode aliasing obscures the physical meaning of the IMF components and affects the subsequent analysis.

The decomposition results of the oil film oscillation signal using BEMD () are shown in Figure 7. The signal z is decomposed into eight complex rotational components, c₁–c₈. After decomposition, the IMF numbers from x are the same as those from y. The noise component c₁ is more easily separated from signal z relative to the case of IMF₁ from signal x. c₂ and c₃ are considered fundamental and oscillatory components, respectively; the frequency component of c₂ is 108 Hz, and the frequency of c₃ is 52 Hz. The two trajectories composed of c₂ and c₃ generated by the BEMD method are shown in Figures 8(a) and 8(b), respectively, while the corresponding ones of IMF₂ and IMF₃ generated by EMD method are shown in Figures 8(c) and 8(d), respectively. Figure 8 indicates that the orbit arrangements obtained by the BEMD method are superior to those of the EMD method.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(a)

(b)

(c)

(d)

Relative to the EMD method, the decomposition effect is improved when using BEMD to decompose the oil film oscillation signal. However, there are still two problems when using BEMD to decompose the oil film oscillation signal. One is the occurrence of modal aliasing in the IMFs, and the other is increased false components. From Figure 6, the modal aliasing occurs at the end of c₄, which causes the value of c₃ to become small at the end. After decomposition, eight IMF components and one residual r are obtained. However, most of the IMF components contain useless information. In most cases, the fault characteristics are often included in higher-energy components. The projection directions and the energy end condition based on the energy threshold are increased in the improved BEMD method to enhance the decomposition quality.

The decomposition results of the oil film oscillation signal are shown in Figure 9 using the improved BEMD method (, λ = 0.05). The oil film oscillation signal is decomposed into four complex rotational components c₁–c₄ and the remaining component r. c₁ is considered as the noise component. According to the Fourier analysis, c₂ is the fundamental signal with a frequency of 108 Hz, and c₃ is the oscillation signal with a frequency of 52 Hz. It is clear that the number of IMFs is reduced, and the fundamental frequency component c₂ and the oscillation component c₃ are more accurately extracted from the original signal. In particular, the value of the c₄ end becomes smaller, and the value of the c₃ end becomes larger, relative to the values of c₃ and c₄ shown in Figure 7. Two trajectories made up of c₂ and c₃ are shown in Figure 10. From c₂ and c₃ three-dimensional time domain and the trajectories, the fundamental frequency signal c₂ has a small elliptic amplitude change. c₃ is a large series of amplitude conversion elliptical compositions that cause a significant oscillation. Relative to the orbit of c₃ in Figure 6, the orbit of c₃ in Figure 10 is improved, particularly in the center region.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(a)

(b)

Increasing the number of signal projection directions results in an increase in the number of projection signals. Then, the tangent mean, which is obtained by interpolating the local maximum of the projected signals with a spline interpolation, is more accurate. However, it is meaningless to continue to increase the number of projection directions when the tangent mean is accurately fitted. It is found that the signal decomposition results of are almost the same as when considering the complex rotation components separated by the original signal. However, the calculation time of the BEMD algorithm with is greatly increased.

The HT is applied to the real and imaginary parts of c₂ and c₃, respectively, and the instantaneous frequency and instantaneous amplitude are obtained, as shown in Figure 11, where a_x2 and a_x3 represent the instantaneous amplitude of the real parts of c₂ and c₃, respectively, a_y2 and a_y3 represent the instantaneous amplitude of the imaginary parts of c₂ and c₃, respectively, f_x2 and f_x3 represent the instantaneous frequency of the real parts of c₂ and c₃, respectively, and f_y2 and f_y3 represent the instantaneous frequency of the imaginary parts of c₂ and c₃, respectively.

(a)

(b)

(c)

(d)

In Figure 11, a_x3 is much larger than a_y3, and their phases are separated by nearly 180°, which shows that the amplitude and the phase of the oil film oscillation signal in different directions can vary. f_x2 and f_y2, and f_x3 and f_y3, are approximately the same. a_x2 and a_y2 show little change, and their phases are the same. Since BEMD is a bivariate extension of EMD, like EMD, BEMD also has an endpoint effect. There are some fluctuations in the instantaneous amplitude and instantaneous frequency due to the end effect and the edge effect of the Hilbert transform. The three-dimensional time domain of a_x2 and a_y2 and of a_x3 and a_y3 is shown in Figure 12. The amplitude range of c₃ is much larger than that of c₂. It is can be inferred that the main component of oscillation is c₃ with the frequency of 52 Hz in the oil film oscillation signal.

(a)

(b)

The decomposition results of the oil film oscillation signal are shown in Figure 13 using the CLMD method proposed in reference [14]. The oil film oscillation signal is decomposed into four complex product functions cpf₁–cpf₄. The noise component is not separated from the oil film oscillation signal using the CLMD method. Moreover, the single component fundamental frequency signal and the oil film oscillation signal were not successfully separated. One of the possible reasons is that in the CLMD algorithm, the complex signal is only projected onto the x-axis and the y-axis, unlike BEMD, which projected on multiple directions. In addition, CLMD is a bivariate extension of LMD. LMD used a moving average algorithm when fitting the signal envelope, which can filter noise to a certain extent. The signals other than the noise component c₁ in Figure 9 are added to obtain a filtered oil film oscillation signal, which is then decomposed by the CLMD method, and the first two decomposed results are shown in Figure 13. It is seen that the single component fundamental frequency signal and the single component oil film oscillation signal are separated.

(a)

(b)

(c)

(d)

cpf₁ and cpf₂ consisted of real part signals and imaginary part signals, both of which were composed of the product of the envelope signal and the pure frequency modulation function. The envelope signal is the instantaneous amplitude of the signal. The corresponding instantaneous frequency was obtained by deriving the inverse function of the cosine pure frequency modulation function. The instantaneous amplitude and instantaneous frequency curves are shown in Figure 14. The instantaneous amplitude and frequency obtained by the CLMD method are smoother than in Figure 12. The reason is mainly that the CLMD method uses the moving average filtering algorithm to obtain the signal envelope curve. However, this is the result of using the CLMD algorithm after filtering out noise with BEMD. If the BEMD algorithm is not used for filtering noise, the instantaneous amplitude and instantaneous frequency curves of the single component were not obtained by the CLMD method.

(a)

(b)

(c)

(d)

4.2. Analysis of Oil Whirl Signal Based on the Improved BEMD Method

In the method similar to that presented in Section 4.1, the oil whirl signals of the rotor test rig with a speed parameter of 4,320 rpm are collected by two orthogonal sensors, as shown in Figure 15. The figure shows the typical whirl phenomenon of large circles with embedded smaller ones. The decomposition results based on the improved BEMD method (, λ = 0.05) are shown in Figure 16.

(a)

(b)

(c)

(d)

(a)

(b)

(c)

(d)

(e)

(f)

Only three IMFs appear in Figure 16, and the single components c₂ and c₃ and the noise component c₁ are successfully separated from the original signal. The HT is applied to the real and imaginary parts of c₂ and c₃, respectively, and the instantaneous frequency and instantaneous amplitude are obtained, as shown in Figure 17. The three-dimensional time domain of a_x2 and a_y2 and of a_x3 and a_y3 is shown in Figure 18. Figure 17 indicates that the frequency of c₂ is approximately twice the frequency of c₃ and that a_x2 is larger than a_y2. In addition, a_y3 is slightly larger than a_x3, but the range of change for a_x3 is greater than the range for a_y3. It is inferred that c₂ is the fundamental frequency signal and that c₃ is the half-frequency signal in the oil whirl signal.

(a)

(b)

(c)

(d)

(a)

(b)

4.3. Analysis of Looseness and Rotor Rubbing Composite Fault Signal Based on the Improved BEMD Method

Loose and rotor rubbing composite faults are set on the test equipment shown in Figure 3 in Section 4.1. Loose fault is set on the nondrive end of the motor, and the distance between the plastic rod and the shaft is fixed near the sensor on the left side of the disk. As the rotor speed increases, the vibration increases and the rubbing fault occurs, which is stable at around 1,700 r/min. The composite fault signals are collected by two orthogonal sensors, as shown in Figure 19. The decomposition results based on the improved BEMD method are shown in Figure 20.

(a)

(b)

(c)

(d)

(a)

(b)

(c)

(d)

(e)

Figures 19(c) and 19(d) indicate that the signal component mainly contain 1X, 2X (X = 28 Hz), and a frequency modulated signal generated due to time-varying stiffness. This phenomenon is similar to that described reference [20]. Four IMFs appear in Figure 20, and the single components 1X, 2X signals and the FM signal c₂ are successfully separated from the original signal. The HT is applied to the real and imaginary parts of c_2,c₃, and c₄, respectively, and the instantaneous frequency and instantaneous amplitude are obtained, as shown in Figure 21. There are some fluctuations in the frequencies of c₂ and c₃, but these fluctuations are different from the random fluctuations in the above cases. They have obvious regularity and are characteristic of FM signals. The frequency modulation characteristics of c₂ are more obvious than those of c₃. In addition, by observing the instantaneous amplitude of c₂, it is seen that c₂ is still an amplitude modulation signal.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 21

The instantaneous amplitude and frequency of c₂, c₃, and c₄ obtained by the HT: (a) the instantaneous amplitude of the real part and imaginary part of c₂; (b) the instantaneous frequency of the real part and imaginary part of c₂; (c) the instantaneous amplitude of the real part and imaginary part of c₃; (d) the instantaneous frequency of the real part and imaginary part of c₃; (e) the instantaneous amplitude of the real part and imaginary part of c₄; (f) the instantaneous frequency of the real part and imaginary part of c₄.

In order to further verify the correctness of the instantaneous amplitude-frequency characteristics of the proposed method, the real and imaginary parts of the composite fault signal z are analyzed separately using synchrosqueezed wavelet transforms (SWT) proposed in reference [21]. The results are shown in Figure 22. It is seen that the time-frequency representations of the composite fault signal z also include the AM-FM signal and the 1X signal, which proves the correctness of the proposed method. Compared with the SWT method, the instantaneous amplitude-frequency characteristics acquired by the HT method are relatively straightforward.

(a)

(b)

4.4. The Bistable Behavior Analysis of the Fan Rotor Based on BEMD

The bistability of the rotor is a nonlinear behavior of the rotor-bearing system, which is the state in which the rotor jumps from one stable state to another, forming a step. The bivariate signal of the bistable behavior is composed of two signals collected by two displacement sensors from orthogonal locations on the experimental devices in literature [22], as shown in Figure 23. Literature [22] shows that the cause of the bistable behavior remains to be further explored. This paper uses this case to illustrate the feasibility of BEMD to analyze nonstationary signals.

The x and y signals in the horizontal and vertical directions of the left and right bearings, respectively, from the fan rotors are collected with four displacement sensors. Letting z = x + jy, the time and frequency domain plots of z are shown in Figure 24, where the fan rotor speed is 5500 rpm, the sampling frequency is 2000 Hz, and the number of sampling points is 1024. The left and right columns, respectively, show the time and frequency domain plots of the vibration signals from the left and right bearings of the fan rotor. Bistable behavior arises in the fan rotor, and the amplitudes of the vibration signals vary significantly in different positions and directions. Further studies are required to explain the causes of this bistability. The present study focuses on extracting the bistable behavioral signal characteristics to verify the feasibility of the proposed method.

The decomposition results of the bistable behavioral signals based on the improved BEMD method are shown in Figure 25. c₁, c₂, c₃, and r are separated in order from z using the improved BEMD method. c₁ shows a random arrangement and is considered the high-frequency noise signal. c₂ is considered to represent the extracted bistable behavior signals. The HT is applied to the real and imaginary parts of c₂ to obtain the instantaneous amplitude and frequency of c₂ from the left and right columns from Figure 25, as shown in Figure 26. Figure 27 shows the three-dimensional time domain of a_x2 and a_y2 from the left and right columns, respectively. Figure 26 shows that the vibration signal amplitude on the left side of the fan decreases from large to small, opposite of the behavior of the right. The horizontal vibration signal amplitude on the left side of the fan is larger than that of the vertical direction signal, opposite of the right. This result validates that the vibration signals from different directions or positions are different when the fan produces bistable behavior. In addition, the time of the bistable behavior can be determined according to the jump point of the amplitude or frequency.

(a)

(b)

5. Discussion

The BEMD algorithm decomposes two orthogonal directions of vibration signals as a complex signal, which is a two-dimensional digital signal processing method, thus ensuring that the real and imaginary parts have the same decomposition scale. Similar to EMD, the envelope mean is critical for the decomposition effect of BEMD, but the envelope mean in BEMD is three-dimensional. If the number of projection directions of the complex signal in three-dimensional space is larger, the corresponding envelope signal is also more. Thus, the envelope mean value is more accurate and the BEMD decomposition effect is better. Increasing the number of projection directions can improve modal aliasing. Like EMD, BEMD also produces false components when decomposing signals. Generally speaking, the energy of the false components is low, and these low-energy false components do not contain fault characteristic information, and the introduction of the energy threshold criterion in the termination condition can increase the decomposition speed of the BEMD.

The experimental results show that there is a certain difference in the existence of vibration signal characteristics in different directions when rotating machinery fails. In addition, when the number of projection directions is increased, the decomposition speed of BEMD will decrease.

6. Conclusions

We use BEMD and HT to extract the instantaneous amplitude-frequency features of rotor faults. A bivariate instantaneous feature extraction method based on the improved BEMD method and the HT is investigated, which extends the fault feature extraction technology to two dimensions. The BEMD method is suitable to analyze the complex, multicomponent bivariate signals. The main single-component bivariate signals are separated from the multicomponent bivariate signals of the fan rotor bistability for the oil film oscillation and the oil film vortex using the improved BEMD method. For the single-component bivariate signal, the HT is used to obtain the corresponding instantaneous amplitude and frequency characteristics. The proposed method can examine the detailed information of a single rotation component.

Data Availability

The data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

Authors’ Contributions

All the authors contributed to this work. Chuanjin Huang conceived and designed the simulation and experiments and drafted the manuscript; Haijun Song performed the simulations and experiments and analyzed the data; and Wenping Lei and Yajun Meng performed the experiments and analyzed the data. All the authors contributed to the writing and discussion of the paper.

Acknowledgments

This research was funded by the Henan Provincial Higher Education Key Research Project (Grant nos. 18A460006 and 19A460029), Henan High-Level Innovative Scientific and Technological Talent Team Construction Project (Grant no. C20150034), and Zhengzhou Institute of Technology Innovation Team Project (Grant no. CXTD2017K1).

References

R. Yan, R. X. Gao, and X. Chen, “Wavelets for fault diagnosis of rotary machines: a review with applications,” Signal Processing, vol. 96, pp. 1–15, 2014.
View at: Publisher Site | Google Scholar
J. Cheng, D. Yu, J. Tang, and Y. Yang, “Application of frequency family separation method based upon EMD and local Hilbert energy spectrum method to gear fault diagnosis,” Mechanism and Machine Theory, vol. 43, no. 6, pp. 712–723, 2008.
View at: Publisher Site | Google Scholar
H. Liu and M. Han, “A fault diagnosis method based on local mean decomposition and multi-scale entropy for roller bearings,” Mechanism and Machine Theory, vol. 75, pp. 67–78, 2014.
View at: Publisher Site | Google Scholar
Z. Zheng, W. Jiang, Z. Wang, Y. Zhu, and K. Yang, “Gear fault diagnosis method based on local mean decomposition and generalized morphological fractal dimensions,” Mechanism and Machine Theory, vol. 91, pp. 151–167, 2015.
View at: Publisher Site | Google Scholar
W. Yang, R. Court, P. J. Tavner, and C. J. Crabtree, “Bivariate empirical mode decomposition and its contribution to wind turbine condition monitoring,” Journal of Sound and Vibration, vol. 330, no. 15, pp. 3766–3782, 2011.
View at: Publisher Site | Google Scholar
L. Qu, X. Liu, G. Peyronne, and Y. Chen, “The holospectrum: a new method for rotor surveillance and diagnosis,” Mechanical Systems & Signal Processing, vol. 3, no. 3, pp. 255–267, 1989.
View at: Publisher Site | Google Scholar
F. Q. Wu and G. Meng, “Compound rub malfunctions feature extraction based on full-spectrum cascade analysis and SVM,” Mechanical Systems and Signal Processing, vol. 20, no. 8, pp. 2007–2021, 2006.
View at: Publisher Site | Google Scholar
Y. Chen, Q. Gao, and Z. Guan, “Self-loosening failure analysis of bolt joints under vibration considering the tightening process,” Shock and Vibration, vol. 2017, Article ID 2038421, 15 pages, 2017.
View at: Publisher Site | Google Scholar
L. Chen, J. Han, W. Lei, Y. Cui, and Z. Guan, “Full-vector signal acquisition and information fusion for the fault prediction,” International Journal of Rotating Machinery, vol. 2016, Article ID 5980802, 7 pages, 2016.
View at: Publisher Site | Google Scholar
C. Chen, Y. Meng, and Y. Du, “Application of the full vector spectrum based on EMD in fault diagnosis of bearings,” Journal of Mechanical Strength, vol. 37, pp. 806–811, 2015.
View at: Google Scholar
C. Huang, X. Wu, and W. Cao, “LMD-based on full vector envelope technique and its application in TRT vibration fault diagnosis,” Electric Power Automation Equipment, vol. 35, pp. 168–174, 2015, in Chinese.
View at: Google Scholar
X. Zhao, T. H. Patel, and M. J. Zuo, “Multivariate EMD and full spectrum based condition monitoring for rotating machinery,” Mechanical Systems and Signal Processing, vol. 27, pp. 712–728, 2012.
View at: Publisher Site | Google Scholar
G. Rilling, P. Flandrin, P. Gonalves, and J. M. Lilly, “Bivariate empirical mode decomposition,” IEEE Signal Processing Letters, vol. 14, no. 12, pp. 936–939, 2007.
View at: Publisher Site | Google Scholar
C. Park, D. Looney, M. M. Van Hulle, and D. P. Mandic, “The complex local mean decomposition,” Neurocomputing, vol. 74, no. 6, pp. 867–875, 2011.
View at: Publisher Site | Google Scholar
N. Rehman and D. P. Mandic, “Empirical mode decomposition for trivariate signals,” IEEE Transactions on Signal Processing, vol. 58, no. 3, pp. 1059–1068, 2010.
View at: Publisher Site | Google Scholar
N. Rehman and D. P. Mandic, “Multivariate empirical mode decomposition,” Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, vol. 466, no. 2117, pp. 1291–1302, 2010.
View at: Publisher Site | Google Scholar
Y. Lv, R. Yuan, and G. Song, “Multivariate empirical mode decomposition and its application to fault diagnosis of rolling bearing,” Mechanical Systems and Signal Processing, vol. 81, pp. 219–234, 2016.
View at: Publisher Site | Google Scholar
C. Huang, Y. Meng, and W. Lei, “Full vector envelope technique based on complex local mean decomposition and its application in fault feature extraction for rotor system,” Journal of Mechanical Engineering, vol. 52, no. 7, p. 69, 2016, in Chinese.
View at: Publisher Site | Google Scholar
G. Rilling, P. Flandrin, and P. Goncalves, “On empirical mode decomposition and its algorithms,” in Proceedings of IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing, pp. 8–11, IEEE, Trieste, Italy, June 2003.
View at: Google Scholar
L. Yang, X. Chen, and S. Wang, “Mechanism of fast time-varying vibration for rotor–stator contact system: with application to fault diagnosis,” Journal of Vibration and Acoustics, vol. 140, no. 1, Article ID 014501, 2018.
View at: Publisher Site | Google Scholar
I. Daubechies, J. Lu, and H.-T. Wu, “Synchrosqueezed wavelet transforms: an empirical mode decomposition-like tool,” Applied and Computational Harmonic Analysis, vol. 30, no. 2, pp. 243–261, 2011.
View at: Publisher Site | Google Scholar
L.-S. Qu, Holospectrum and Holobalancing Technique in Machinery Diagnosis, Beijing Science Press, Beijing, China, 2007.

Copyright

Copyright © 2019 Chuanjin Huang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

2582

Downloads

936

Citations