On-line vibration monitoring of bearing faults in induction machine

47 , 09010 (2016) MATEC Web of Conferences 83

DOI: 10.1051/ matecconf/20168309010

CSNDD 2016

On-line On -line vibration monitoring of bearing faults in induction machine using Cyclic Spectral Analysis F. Tafinine Laboratoire LTII, Faculté de Technologie, Université A.Mira de Béjaia , 06000 Bejaia, Algérie

Abstract. A robust signal processing method for developing an on-line bearing fault detection based on Cyclic Spectral Analysis is presented in this paper. The exploitation of cyclostationarity has recently been proved extremely fruitful to conceive high accuracy diagnostics tools. Cyclostationarity encompasses a subclass of nonstationary signals which exhibit some cyclical behaviour; it ideally fits the property of many rotating and reciprocating machine vibrations, due to the inherent periodic modulations that these sustain during operation. The traditional spectral analysis is not dedicated for non-stationary signals and for an on-line diagnosis. The proposed technique is tested with different bearing conditions, the results show that this technique is especially useful for non-stationary features extraction; it gives a good basis for an integrated induction motor condition monitor.

1 Introduction Asynchronous machines are far more widely used in the industrial world for their simplicity, their robustness and their cost. The importance of detection and diagnosis of faults in machines have expanded considerably due to the increased complexity of plant equipment and high costs associated with failure and shutdown. Faults detection and isolation problems can be solved by different techniques: model based, signal based or knowledge based methods. Faults can be due to electrical or mechanical failures. The most common failure in an asynchronous machine is bearing failure [1]. Vibration analysis techniques were used for the monitoring and diagnosis of huge rotating machines. Different methods were used. Time domain statistical method such as energy monitoring (energy increases as the machine’s condition deteriorate), frequency domain method such as spectrum (characteristic frequencies associated with the failure appear in the spectrum), parametric models (the parameters of the model change with the machine’s condition). Vibrations can be measured by attaching displacement or velocity transducers, and thus requires direct access to the machine. In recent years much work has been carried out on the investigation of possible fault operations of electrical machines. The usual methods for detection of faults are based on vibration analysis [3-6,10-11] or on Motor current signature analysis (MCSA) [1- 2, 12-13]. The bearings condition monitoring has received considerable attention for many years [3, 9, 14]. Many methods are available for detecting faults in bearing

systems. The majority of these methods assume that faults in the bearings produce impulses [10, 11]. The Fourier analysis may be the most commonly used approach for bearing fault diagnosis. However, it is not appropriate because: first, vibration signal is generally non-stationary [5], and second, this approach cannot be applied for an on-line diagnosis, because the operator must inspect the total vibration signal spectrum in order to detect the frequency of defects. In this paper, we will present the detection of the outer race defect using Cyclic Spectral Analysis. The results obtained confirm the effectiveness of this method in the detection of the frequency of motor bearing damage. A non-invasive technique is now proposed. It is very practical and can be implemented at low cost. This paper is organized as follows: Section 2 presents experimental setup and the bearing faults characteristic frequencies. Section 3 reviews the theoretical background of the proposed diagnostic procedure. First we introduce the principles of the spectral domain analysis using Fourier transform. We emphasize on how the relevant information can be extracted from the spectral density using the FFT algorithm. Second, we describe the Cyclic Spectral Analysis and how it can be used for diagnosis by extracting the relevant frequencies characterizing the motor operating mode. Finally, a comparative approach of the proposed diagnostic procedure and several usual vibratory diagnostic methods are presented in section 4. And the conclusions are given in section 5.

© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (http://creativecommons.org/licenses/by/4.0/).


DOI: 10.1051/ matecconf/20168309010

CSNDD 2016

where « BD » is the ball diameter, « PD » is the pitch diameter, « β » is the contact angle of the balls on the race

2 Experimental setup We have used a motor of 380 V; 50 Hz; 1.1 kW; number of pair poles (P=2). The experiments are realized with several levels of load and speed. Tests are conducted for a healthy bearing then with an outer race fault. The signals are measured using multiple vibration sensors installed on the induction machine. Note that the all measurements are taken at stationary regime.

« N » is the number of balls. 2.2 Frequency characteristics of failure The Benchmark is equipped with an RS bearing type 6028-2RS. The defect frequency at the speed of 3004 rpm is 4.58 Hz (see table 1). Table 1. Frequency characteristics of the bearing type 60282RS.

Frequency (Hz) Speed (rpm) 3004

fc

fb

fo

fi

0.28

5.43

4.58

5.81

3 Applications and results Fig. 1. Scheme of the experimental setup

3.1. Classical spectral analysis 2.1 Bearing failure The vibrations frequencies depend: on the rotation speed “ f r ”, on the geometrical parameters of the bearing (figure 2) and on the type of defects [2, 9]. The vibration frequencies of bearing failure are given in the equations bellow [7, 8]:

the psd (db)

The classical method converts the sampled signal to the frequency domain using a FFT (Fast Fourier Transform) algorithm. The generated spectrum includes only the magnitude information about each frequency component. Signal noise that is present in the calculated spectrum is reduced by averaging a predetermined number of generated spectra. The power spectral density is illustrated in the figure 3. The result shows a periodic modulation of 4.6 Hz (sideband spacing = 4.6 Hz) in the defect machine spectrum. This modulation is due to the outer race defect.

Fig. 2. Geometrical parameters of the bearings.

Outer race bearing fault frequency F0: N BD ⎡ ⎤ Fo= cos β ⎥ f r ⎢1 − 2 PD ⎣ ⎦ Inner race bearing fault frequency Fi : N BD ⎡ ⎤ Fi= cos β ⎥ f r ⎢1 + 2 PD ⎣ ⎦ Cage defect frequency Fc: 1 BD ⎡ ⎤ Fc= f r ⎢1 − cos β ⎥ 2 PD ⎣ ⎦ Ball defect frequency Fb: Fb=

2 ⎡ ⎛ BD PD ⎞ ⎤ f r ⎢1 − ⎜ cos β ⎟ ⎥ 2 BD ⎢⎣ ⎝ PD ⎠ ⎥⎦

Frequency (Hz)

(1)

Fig. 3. PSD of the vibration signal with outer race defect.

(2) 3.2 Cyclical spectral analysis

(3)

The cyclic power spectrum. S x ( f ; α ) of signal x(t), is defined as [17], [18]:

(4)

* 1 ⎧⎪ ⎛ α⎞ ⎛ α ⎞ ⎫⎪ S x ( f ;α ) = lim Ε⎨ X l ⎜ f + ⎟ X L ⎜ f − ⎟ ⎬ (5) L →∞ L 2⎠ ⎝ 2 ⎠ ⎪⎭ ⎪⎩ ⎝

2


DOI: 10.1051/ matecconf/20168309010

CSNDD 2016

4 Comparison of the two analysis tools where X L (f ) is the Fourier transform of x(t) limited in the interval L. The definition of the cyclic coherence is [15], [16]: 2

γ x ( f ;α ) =

S x ( f ;α )

Vibrations produced by faulty rolling-element bearing are essentially random cyclostationary [18]. In order to detect the frequency of defects it is habitual to demodulate the signal in different frequency bands in the total frequency axis. This method is the classical technique in the spectral analysis (PSD). The Cyclic Spectral Analysis can replace this empirical process by a precise methodology to quickly find the frequency of the faults. The new diagnosis procedure overcomes the disadvantages related to motor parameters dependency. In contrast to the usual vibrations signal spectral analysis, the proposed diagnosis method allows also the ability to extract quickly the characteristic frequencies that distinguish the motor operating modes. The cyclic coherence is actually the cyclic spectral quantity that offers the best fault detection tool, it reveals the discrete spectral signature of the fault. Its magnitude serves as a relative measure of the fault severity. These observations have been supported by experimental results.

2

α⎞ ⎛ α⎞ ⎛ Sx ⎜ f + ⎟ Sx ⎜ f − ⎟ 2⎠ ⎝ 2⎠ ⎝

(6)

This expression presents a correlation coefficient between two components at frequencies f + α / 2 and f − α / 2 . It takes values between 0 and 1, where 0 expresses nonlinear correlation, and 1 in the case of a linear relationship between the two components [17]. The diagnostics information can be revealed in the α -frequency domain, the magnitude of the cyclic coherence can be used as a measure of the fault severity [19]. Cyclic coherence of the damaged bearings (figure 4) illustrates significant cyclostationary. A fundamental spectral line at α = 4.5 Hz without sidebands is, according to table 1, the characterization of an outer-race fault. In the cyclic coherence, some bands located within [0; 3] Hz might be due to structural resonances that are excited by the impacts of the rolling elements on the faults (ie. Fc, fr). These observations agree with the frequency band in the power spectral densities in figure 3.

5 Conclusions This paper has investigated a method for detecting bearing faults using the cyclical spectral analysis. Machine fault features extraction without prior information of the fault type is important for diagnosis and maintenance planning. The Cyclic Spectral Analysis technique is qualified as a parsimonious method for exploring the information in the entire spectrum of the vibrations signal in the cyclic frequency domain. The resulting procedure is fined suitable for an on-line diagnosis, without manual research of the frequency band containing the fault.

(a)

Acknowledgment fr

Fo

(b)

The author wishes to express his gratefulness to the members of the Roberval Laboratory (UTC France) for their support in the experimental setup.

Fc

References 1. R.R. Schoen, T. Ha betler, F. Karman, R. Bartheld, IEEE Trans . Ind. App. 31, 1274 (1995) 2. M.E.H. Benbouzid, IEEE Trans . Ind. Electron. 47, 984 (2000) 3. R. Boustany, J. Antoni, in Proc. of the 16th IFAC World Congress, Prague, Czech Republic, (2005) 4. F. Zhipeng, C. Fulei, Shock and Vib. 1155 (2015) 5. Y. Yang, D. Yu, J. Cheng, Measurement, 40 943 (2007) 6. R. Boustany, J. Antoni, Mech. Syst. Sig. Process. 19, 1245 (2005)

Fig. 4 Estimated cyclic spectral coherence (a) and integrated spectral coherence (b) of the vibration signal with an outer-race fault.

3


DOI: 10.1051/ matecconf/20168309010

CSNDD 2016

13. M. Blodt, M. Chabert, J. Regnier, J. Faucher, IEEE Trans. Ind. App. 42, 1454 (2006) 14. R. Isermann, Control Eng. Pract. 5, 639 (1997) 15. J. Antoni, F. Bonnardot, A. Raad, M. El Badaoui, Mech. Syst. Sig. Process. 18, 1285 (2004) 16. R.B. Randall, J. Antoni, S. Chobsaard, Mech. Syst. Sig. Process. 15, 945 (2001) 17. A. Raad, J. Antoni, M. Sidahmed, Mech. Syst. Sig. Process. 22, 574 (2008) 18. A.C. McCormick, A.K. Nandi, J. Sound vib. 12, 225 (1998) 19. C. Capdessus, M. Sidahmed, J.L. Lacoume, Mech. Syst. Sig. Process. 14, 371 (2000)

7. B. Raison, G. Rostaing, O. Butscher, C.S. Maroni, in Proc. 28th Annual Conference of the Industrial Electronics Society, Sevilla, Spain, 2002. 8. W.T. Thomson, in Proc. of the IEEE Symposium on Diagnostics for Electrical Machines Power Electronics and Drives (SDEMPED), Gijon, Spain, 1999. 9. T.A. Harris, Rolling bearing analysis, (4th Edition); John Wiley & Sons, Inc, New York, USA, 2001. 10. N.G. Nikolaou, I.A. Antoniadis, Mech. Syst. Sig. Process. 16, 667 (2002) 11 . T.C. Lim, R. Singh, J. Sound vib. 139, 179 (1990) 12. M. Blodt, Condition Monitoring of Mechanical Faults in Variable Speed Induction Motor Drives, Ph.D. Thesis, INP Toulouse, France, (2006)

4