A Review of Vibration Analysis Techniques for Rotating Machines

A Review of Vibration Analysis Techniques for Rotating Machines

Saurabh Singh

Student

Department of Mechanical Engineering Maulana Azad National Institute of Technology Bhopal, India

Dr. Manish Vishwakarma Assistant Professor

Department of Mechanical Engineering Maulana Azad National Institute of Technology Bhopal, India

Abstract The safety, reliability and efficiency of a rotating machine are of a major concern in an industry. Condition monitoring of a machine helps in retaining the efficiency and performance of a machine to its optimum level. The condition monitoring of a rotating machine is efficient, but often complex and labor intensive task for maintenance of the machine. Vibration analysis is a technique used for condition monitoring of the machine. Effective vibration signal extracting techniques have a critical role in efficiently diagnosing a rotating machine. Many vibration signal extracting techniques have been proposed during past some years. The paper presents review of some vibration feature extraction methods applied to different types of rotating machines.

Keywords Time domain, frequency domain, crest factor, fourier transform, wavelet.

1. INTRODUCTION

   Rotating machines are widely used in todays industries. Some of these are very complex and critical for operation. The machine failure may result in costly downtime and loss of men and money. An effective diagnosis system is needed to predict the condition and reliable lead time of the machine. Therefore, an effective and efficient condition monitoring and fault diagnosis system is highly desirable in an industry. Vibration analysis is mostly used for fault detection in rotating machine.

2. VIBRATION ANALYSIS

   It is natural for a machine to vibrate. Even in the best operating conditions machines will have some vibration because of small minor defects. Each machine has a level of vibration which may be regarded as normal. However, sometimes the machine vibration level increases or becomes excessive. Some mechanical trouble is usually the reason. The reasons behind the excessive vibration include- unbalance, misalignment, worn gears or bearings, looseness etc. The vibration level of the machine is measured with the help of sensors. These are proximity sensor, velocity transducer and accelerometer. Accelerometers are mostly used for vibration analysis.

   Fault diagnosis in the rotating machine is conducted in the following steps- data acquisition, feature extraction and fault

   detection. Effective features extraction very critical foe the efficient fault diagnosis system.

   Vibration signals collected by signals are often contaminated by noise signals. These contaminated signals can be unfit for fault diagnosis. Without the assistance of some certain techniques the vibration features or signatures can go undetected. Feature extraction techniques can locate certain components in signals to help detection of machine faults [1]. Various vibration analysis techniques have been applied to the fault diagnosis of the rotating machines. In past some decades, some study has been done in reviewing vibration techniques from a different point of view. In 1980s Mathew and Alfredso presented a review of vibration monitoring techniques in time and frequency domain for rolling element bearings [2]. McFadden, Smith [3] and Kim [4] included classical non-parametric spectral analysis, principal component analysis, joint time frequency analysis and the discrete wavelet transform. Lebold and McClintc [5] reviewed statistical method for extracting features for gearbox fault diagnosis. Tendon and Chaudhary [6] reviewed vibration and acoustic measurement techniques for rolling element bearing fault analysis. Tashin Doguer and Jens Strackeljan [7] reviewed time domain feature extraction for diagnosing small roller bearing defects.

   Fig- Overview of Vibration Analysis

   In this paper a review of vibration analysis techniques for a variety of rotating machines is presented. These techniques are grouped in three categories, time domain, frequency domain and time-frequency domain as shown in figure. In each category diverse range of techniques are presented, some of which are based on common principles.

   1. Time domain

      Vibration signals are obtained as a series of values representing proximity, velocity or acceleration in the time domain. This section presents recent study on vibration techniques in the time domain. These techniques are categorized in the following groups as shown in the table 1-

      Table 1- Time Domain Vibration Analysis Techniques

      Statistical parameters- these include Root Mean Square (RMS), Mean, Variance, Skewness, Kurtosis and Crest factor.

      Root mean square value and crest factor has been applied in diagnosing bearings and gears [6]. The RMS value of a vibration signal is a time domain feature. It presents the power content in the vibration. RMS value is very helpful in detecting imbalance in rotating machinery. In the time domain RMS value is the most basic method used to detect the faults in rotating machine, but this is not capable of detecting faults when the problem is in early stage.

      Crest factor is another measure in the time domain vibration analysis. Crest factor is defined as the ratio of the peak value of the input signal to the RMS value. In time series signal peaks will result in an increase of the crest factor value. In rotating machines tooth breakage or fault in outer race of bearing leads to generation of impulse vibration signals. Here the crest factor will be high so it helps in detecting faults in gear tooth or bearing.

      Statistical analysis is useful in detecting faults in industrial machinery. Tandon [6] review the time domain vibration analysis for detecting faults in bearing. He showed that probability function is correlated with bearing defects. The probability density of acceleration of a bearing in good condition has a Gausian distribution, whereas for a defected bearing the probability density distribution is non-Gausian and with dominant tail. This is because of increase in no. of high acceleration signals. Mathew and Alfredson also showed non-Gausian distribution for defected bearing. Andrade[8] compared the Cumulative Density Function (CDF) of a

      vibration signal with the CDF of a reference distribution to detect gear tooth fatigue crack.

      Statistical moments are also helpful in vibration analysis. Mean, variance and skewness are first, second and third moment of probability distribution respectively. Kurtosis is the fourth moment of probability distribution. It measures the flatness of a distribution to a normal distribution.

      The root mean square, peak value, kurtosis and crest factor have been combined with high frequency resonance techniques to detect damage in roller bearings [10].

      Time Synchronous averaging (TSA) method- It includes Time Synchronous Average (TSA) signal, residual signal (RES) and difference signal (DIFS).

      Time synchronous averaging is a signal processing technique which extracts periodic waveform from noisy data. The TSA signals are obtained by time synchronous averaging of primary vibration signals and reducing the redundant noise. TSA allows the vibration of the gear under analysis to be separate from other rolling components. It separates the noise sources from the vibration signals which are under watch. The signals after TSA can provide the information related to the faults. Synchronous averaged signals have been utilized for fault diagnosis rolling element bearings and gears [10-11]. Eric Bechhoefer [12] reviewed the TSA techniqu and showed that it has similar performance in time and frequency domain.

      Residual signal (RES) was used by Lebold [5] for gear diagnosis. He used RES consisted of the time synchronous averaged signal with the primary meshing and shaft components along with their harmonics removed. Difference signals (DIF) are calculated by removing the regular meshing components from the time synchronous averaging signals. The DIF was used for gear fault diagnosis effectively [13].

      Filter based methods- these methods include demodulation, prony method and adaptive noise cancelling (ANC).

      Filters are used in feature extracting techniques for removing contaminated noise and isolated signals from raw signals. These methods are generally called filter based methods. Demodulation, prony model and adaptive noise cancelling are filter based methods.

      Demodulation, which includes phase and amplitude demodulation, is a widely used signal processing technique. The amplitude demodulation is also called envelop, or resonance demodulation, or high frequency demodulation technique. Amplitude demodulation is a process which separates low level low frequency signal from background noise [14]. This enables the signal to be easily measured. Amplitude demodulation has been applied successfully in the diagnosis of gears [15] and bearings [16] faults. The phase demodulation emphasized the band associated with the structural resonance excited by the fault induced impacts [14].

      Demodulation uses a conventional Infinite Impulse Response (IIR) filters such as Butterworth, Chebyshev, Bessel and Elleptic in pass band or stop band. A Prony model based technique was applied by Chen Z [17] to bearing fault diagnosis. Pronys model is an algorithm for finding an IIR filter with a prescribed time domain impulse response. Li Z

      and Fu Y [18] used adaptive filter for diagnosing bearing faults. Adaptive noise cancelling technique has been applied successfully for detecting faults in roller bearing [19]. Noureddine [20] shows that Prony based model is very effective in detecting fault in gear box.

      Stochastic methods- these include chaos, blind deconvolution, thresholding, and autoregressive model based methods.

      Stochastic parameters have been used for vibration analysis in time domain. Chaos and the correlation dimension have been used to identify many induced faults of different severity in a rolling element bearing [21]. The correlation dimension is helpful in classifying different faults intelligently [22]. Nirbito [23] proposed and studied the suitability of blind deconvolution for the enhancement of bearing signals which were corrupted by noise signals. Threshold denoising (which include soft thresholding and hard thresholding) have been used to denoise vibration signals. The thresholding denoising methods can be used by combining it with envelop or some other methods together for rotating machine fault diagnosis. Soft thresholding method and hard thresholding methods are also being used for rotating machine fault diagnosis [25]. The autoregressive model base has been applied by Wang, W [26] in fault diagnosis.

   2. Frequency domain and Time-Frequency domain

   At present frequency domain features and time-frequency domain features are widely used for vibration analysis of rotating machine. Generally frequency domain features are more capable than time domain features in indicating the faults in rotating machine. This is because the resonance frequency component or the fault frequency component can be detected more easily in frequency domain features as compare to time domain features.

   Appropriate vibration techniques need to be selected as per the need to obtain optimal output from vibration analysis. Some developed frequency domain techniques and time- frequency domain techniques are given in the table 2.

   This section includes the Fast Fourier Transformation (FFT) and various time frequency representations and time frequency scale analysis techniques. The table shows Frequency and time-frequency analysis methods which are being currently used for analysis of vibration signals of rotating machines.

   Frequency domain analysis or spectral analysis of vibration signals is most widely utilized technique for bearing fault detection. The Fast Fourier Transform (FFT) is mostly used diagnosis technique to identify the frequency features of vibration signals. These signals may be raw signals or processed signals.

   The power spectrum whose amplitude is the square of the amplitude of the vibration spectrum is more effective method to diagnose fault in rotating machine. The higher order spectrum is also called bispectrum and it can be applied for rotor bearing fault diagnosis [29].

   The bicoherence is a third order vibration spectrum it is used to measure the phase coherence among three spectral components due to non linear wave coupling. It has been for bearing faults diagnosis [30].

   Table 2- Frequency Domain and Time-Frequency Domain Techniques Power cepstrum is the logarithm of the power spectrum. Power cepstrum has been successfully used for rotating machinery fault diagnosis [6].

   Cycloststionary is the second order of a frequency domain synchronised averaging method. The second order cyclostationary is an effective tool used for detecting early faults in gear system. It has been used for early diagnosis in gear system [31]. Bouillant L and Sidahmed [32] compared the cyclostationary and bilinearity and applied cyclostationary for early diagnosis of helicopter gearbox.

   During the past years time-frequency domain vibration analysis techniques have been researched and applied for machine diagnosis. The main feature of these time-frequency techniques is that these are capable of representing vibration signals in both time and frequency domain. This characteristic helps in analyzing vibration signals which are non stationary. Initially time-frequency analysis techniques, windowed Fourier transform [33] and Short Time Fourier Transform (STFT) were used for machinery condition monitoring.

   The Winger distribution [35] and spectrogram [36] are the well known quadric time frequency representations which are used for diagnosing gear faults. The nature of such signals sauses significant interfering crosss-terms which do not permit a straight forward interception of energy distribution. The directional Choi-Williams distribution (dCWD) was proposed for rotating machinery fault diagnosis [37]. Directional Winger distribution (DWD) has been applied in analysing the order of rotating machine [38].

   The third and fourth order Winger moment spectra also called Winger bi-spectra and Winger tri-spectra were also used for rotating machine vibration signals [39]. The fourth order Winger moment spectra has been applied for diagnosis of valves system faults in an engine [40].

   The continuous wavelet transform (CWT) is developed base on Short Time Fourier Transform (STFT) with better time frequency resolution. It has been applied in fault diagnosis of of rotating machine. The scalogram, second order frequency modulus of CWT was applied for gear fault analysis [26].

   The Discrete wavelet transform (DWT) was used to diagnose spalling in ball bearings [42]. Discrete wavelet packet analysis (DWPA) [43] and discrete wavelet analysis [44] have also been used in fault diagnosis. A time frequency scale domain (TFS) is also useful in diagnosis of rotating machine faults.

3. CONCLUSION

Vibration feature extraction techniques have a critical role in fault diagnosis of rotating machine. The feature extracting techniques are improving all the time. Now with increasing use of computer science the vibration analysis is becoming more efficient and effective.

Time domain techniques include raw signals, filter based signals, stochastic and model based methods. The statistical values such as RMS, mean, kurtosis and crest factor are compared with threshold value for fault detection in rotating machines. Time domain vibration signals hav some limitations for detecting early fault generation. So the research has been conducted for increasing sensitivity of statistical parameters. Filter based methods such as modulation are being used effectively which separate fault signals from unwanted signals such as noise.

Frequency domain features are generally more effective in detecting faults as compared to time domain features. Time frequency domain features are useful in diagnosis of non stationary vibration signals. The research is being conducted to increase the order of transformation parameters. Time frequency techniques are also being researched for analysing vibration signals for specific applications. For example the dyadic discrete wavelet transform is used for low frequency bands rather than high frequency bands. Discrete wavelet packet analysis is applicable for both low frequency and high frequency bands, so making it more powerful for rotating machine fault analysis.

Researches are being conducted for automating diagnosis procedures. Tools such as expert systems, neural network, and fuzzy network are being used with above discussed techniques for increasing effectiveness of fault diagnosis.

REFERENCES

1. Yang, Hongyu and Mathew, Joseph and Ma, Lin, Vibration feature extraction techniques for fault diagnosis of rotating machinery : a literature survey, In: Asia-Pacific Vibration Conference,November 2003.

2. Alfredson R J and Mathew J, Frequency domain methods for monitoring the condition rolling element bearings. Mechanical Engineering Transactions – Institution of Engineers Australia,1985. ME 10(2): 108-112.

3. McFadden P D and Smith J D, The vibration produced by multiple point defects in a rolling element bearing. Journal of Sound and Vibration, 1985 , 98(2): 263-273.

4. Kim Y W, et al., Analysis and processing of shaft angular velocity signals in rotating machinery for diagnostic applications. in Acoustics, Speech, and Signal Processing, 1995. ICASSP-95., 1995 . 2971-2974 vol.5 Dept. of Mech. Eng., Ohio State Univ., Columbus.

5. Lebold M, et al., Review of vibration analysis methods for gearbox diagnostics and prognostics, in the 54th meeting of the society for machinery failure prevention technology, 2000. 623-634 Virginia Beach,VA.

6. Tandon N and Choudhury A, A review of vibration and acoustic measurement methods for the detection of defects in rolling element bearings, Tribology International,1999. 32(8): 469-480.

7. Chow M-Y, Guest editorial special section on motor fault detection and diagnosis. Industrial Electronics, IEEE Transactions on, 2000; 47(5): 982-983.

8. Andrade F A, Esat I, and Badi M N M, A new approach to time- domain vibration condition monitoring: gear tooth fatigue crack detection and identification by the kolmogorov smirnov test. Journal of Sound and Vibration,2001, 240(5): 909-919.

9. Williams T, et al., Rolling element bearing diagnostics in run-to-failure lifetime testing. Mechanical Systems and Signal Processing, 2001,15(5): 979-993.

10. Mcfadden P D and Toozhy M M, Application of synchronous averaging to vibration monitoring of rolling element bearings. Mechanical Systems and Signal Processing,2000, 14(6):891-906.

11. Samuel P D and Pines D J, Vibration separation methodology for planetary gear health monitoring. Proceedings of the SPIE The International Society for Optical Engineering,2000, 3985: 250-60.

12. Eric Bechhoefer1, Michael Kingsley, A Review of Time Synchronous Average Algorithms. Annual Conference of the Prognostics and Health Management Society, 2009.

13. McClintic K, et al., Residual and difference feature analysis with transitional gearbox data. in the 54th meeting of the society for machinery failure prevention technology.2000,635-645 Virginia Beach,VA.

14. Wang W Q, Ismail F, and Golnaraghi M F, Assessment of gear damage monitoring techniques using vibration measurements. Mechanical Systems and Signal Processing,2001, 15(5): 905-22.

15. Wang W, Early detection of gear tooth cracking using the resonance demodulation technique. Mechanical Systems and Signal Processing,2001, 15(5): 887-903.

16. Andrzej Klepka, Wavelet Based Signal Demodulation Technique for Bearing Fault Detection. Mechanics and Mechanical Engineering Vol. 15, No. 4 (2011) 6371

17. Chen Z and Mechefske C K, (Diagnosis of machinery fault status using transient vibration signal parameters. JVC/Journal of Vibration and Control, 8(3): 321-335.

18. Li Z and Fu Y, Adaptive noise cancelling technique and bearing fault diagnosis. Journal of Aerospace Power/Hangkong Dongli Xuebao,1990, 5(3): 199-203.

19. Shao Y and Nezu K, Detection of self-aligning roller bearing fault by asynchronous adaptive noise cancelling technology. JSME International Journal, Series C,1999, 42(1): 33-43.

20. Noureddine Haloui, Djamel Chikouche and Messaoud Benidir, Diagnosis of Gear Systems by Spectral Analysis of Vibration Signals. IJCSNS International Journal of Computer Science and Network Security, VOL.7 No.10, October 2007.

21. Logan D, Using the correlation method for the detection of faults in rolling element bearings, Monash University: Melbourne,1996.

22. Mevel L, Hermans L, and Van Der Auweraer H, (1999), Application of a subspace-based fault detection method to industrial structures. Mechanical Systems and Signal Processing, 13(6): 823-838.

23. Nirbito W, C.C.Tan, and Mathew J, The enhancement of bearing signals corrupted by noise using blind deconvolution-a feasibility study. in ACSIM,2000.

24. Wang W J, et al., The application of some non-linear methods in rotating machinery fault diagnosis. Mechanical Systems and Signal Processing,2001, 15(4): 697-705.

25. Lin J and Qu L, Feature extraction based on morlet wavelet and its application for mechanical fault diagnosis. Journal of Sound and Vibration,2000, 234(1): 135-148.7

26. Wang W and Wong A K, Some new signal processing approaches for gear fault diagnosis. in Signal Processing and Its Applications, 1999. ISSPA '99. Proceedings of the Fifth International Symposium on. 587- 590 vol.2 Aeronaut. & Maritime Res. Lab., Defence Sci. & Technol. Organ., Melbourne, Vic., Australia: New Development, Practical.

27. Burgess P F J, (1988), Antifriction bearing fault detection using envelope detection. Transactions of the Institution of Professional Engineers New Zealand, Electrical/Mechanical Chemical Engineering Section,15(2/EMCh): 77-82.

28. McFadden P D and Smith J D, (1984), Vibration monitoring of rolling element bearings by the high-frequency resonance technique-a review. Tribology International,1988, 17(1): 3-10.

29. Yang D-M, et al., Third-order spectral techniques for the diagnosis of motor bearing condition using artificial neural networks. Mechanical Systems and Signal Processing,2002, 16(2-3): 391-411.

30. Li C J, Ma J, and Hwang B, Bearing condition monitoring by pattern recognition based on bicoherence analysis of vibrations. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science,1996, 210(3): 277-285.

31. Capdessus C, Sidahmed M, and Lacoume J L, Cyclostationary processes: application in gear faults early diagnosis. Mechanical Systems and Signal Processing,2000, 14(3): 371-385.

32. Bouillaut L and Sidahmed M, (Cyclostationary approach and bilinear approach: Comparison, applications to early diagnosis for helicopter gearbox and classification method based on hocs. Mechanical Systems and Signal Processing, 15(5): 923-943.

33. Pan M-C, Sas P, and van Brussel H, Nonstationary time-frequency analysis for machine Condition monitoring. in Time-Frequency and Time-Scale Analysis, 1996., Proceeding of the IEEE-SP International Symposium on. 477-480 Dept. of Mech. Eng., Katholieke Univ., Leuven, Belgium.

34. Klein R, Ingman D, and Baun S, Non-stationary signals: phase-energy approach-theory and simulations. Mechanical Systems and Signal Processing, 2001, 15(6): 1061-1089.

35. In Soo K and Whan Woo K, The development of reactor coolant pump vibration monitoring and a diagnostic system in the nuclear power plant. ISA Transactions, 2000, 39(3): 309 16.

36. Wang W J and McFadden P D, Early detection of gear failure by vibration analysis i. calculation of the time-frequency distribution. Mechanical Systems and Signal Processing, 1993, 7(3): 193-203.

37. Lee S u, Robb D, and Besant C, The directional choi-willams distribution for the analysis of ro torvibration signals. Mechanical Systems and Signal Processing, 2001, 15(4): 789-811.

38. Han Y-S and Lee C-W, Directional Wigner distribution for order analysis in rotating/reciprocating machines. Mechanical Systems and Signal Processing, 1999, 13(5): 723-738.

39. Lee S K and White P R, Higher-order time-frequency analysis and its application to fault detection in rotating machinery. Mechanical Systems & Signal Processing, 1997, 11(4): 637-650.

40. Lee S K, (A new method for smoothing non-oscillation cross-terms in sliced Wigner fourth-order Moment spectra. Mechanical Systems and Signal Processing, 15(5): 1023-1029.

41. Samuel P and Pines D, Health monitoring/damage detection of a rotorcraft planetary geartrain system using piezoelectric sensors. Proceedings of the SPIE The International Society for Optical Engineering, 1997, 3041: 44-53.

42. Mori K, et al., Prediction of spalling on a ball bearing by applying the discrete wavelet transform to vibration signals. Wear, 1996, 195(1-2): 162-168.

43. Nikolaou N G and Antoniadis I A, Rolling element bearing fault diagnosis using wavelet packets. NDT & E International, 2002, 35(3): 197-205.

44. Goumas S, et al., Intelligent on-line quality control of washing machines using discrete wavelet analysis features and likelihood classification. Engineering Applications of Artificial Intelligence, 2001, 14(5): 655-666.