Statistic-based spectral indicator for bearing fault detection in permanent-magnet synchronous machines using the stator current

https://doi.org/10.1016/j.ymssp.2014.01.006Get rights and content

Highlights

  • An indicator based on the statistical analysis of the current spectrum is proposed.

  • It was evaluated on two faulty bearings for different speeds with a rigorous metric.

  • A fault threshold evaluation was performed for four different speeds.

  • The proposed indicator strongly correlates with the vibration indicator.

  • The proposed approach is more robust than the classical energy-based approach.

Abstract

In this paper, an original method for bearing fault detection in high speed synchronous machines is presented. This method is based on the statistical process of Welch׳s periodogram of the stator currents in order to obtain stable and normalized fault indicators. The principle of the method is to statistically compare the current spectrum to a healthy reference so as to quantify the changes over the time. A statistic-based indicator is then constructed by monitoring specific harmonic family. The proposed method was tested on two experimental test campaigns for four different speeds and compared to a vibration indicator. The method was evaluated using a rigorous performance evaluation metric. A threshold evaluation was performed and shows that the proposed method is very tolerant to the machine speed. Thus, the use of a unique fault threshold whatever the speed can be considered. Results showed excellent agreement as compared with the vibration indicator, with an overall correlation of r=0.74 and only 4% of false alarms. Performance demonstrated by this novel method was superior to those of a classical energy-based indicator in terms of correlation with the vibration indicator and detection stability. Moreover, results also showed a better robustness of the proposed method since good performance can be obtained with the same detection threshold whatever the speed or the measure campaign whereas it needs to be redefined for each case with the classical indicator. This work shows the advantages of a statistic-based approach in order to increase the robustness of bearing fault detection in permanent-magnet synchronous machines.

Introduction

Over the past few years, electrical machines are more and more used in many industrial applications. So, monitoring has become an important industrial research area in order to assess their safety and reliability. There are a lot of different causes for failures in electrical machines such as eccentricity, load torque oscillations, and stator turn short circuit. A review of these different causes can be found in [1]. Among them, it has been shown in [2] that ball bearing defects are responsible for 40% of machine failures. Bearing faults can lead to critical events such as abnormal temperature or vibration level, rotor locking, or stator friction. Therefore, manufacturers show more and more interests in monitoring their health state in order to guarantee the availability and the predictive maintenance in actuator elements like machine for example.

Bearing faults are part of mechanical defaults. So, the first approach used to monitor bearing faults has been to analyze the vibrations of electrical machines. The traditional technique is to monitor characteristic frequencies of specific bearing damages in the vibration spectrum as presented in [3], [4], [5]. The spectrum is generally computed using the Fast Fourier Transform (FFT). These techniques imply however to know precisely the resonance frequencies for bearing faults and a mathematical of model of the machine is often needed to identify them. This point can be partly solved using the Spectral Kurtosis in order to detect the frequency bands with the maximum impulsivity [6]. Other representation modes have been proposed such as wavelet [7] or Empirical Mode Decomposition [8] in order to detect changes in the vibration spectral content. The statistical analysis of the spectral content has also been proven to be very efficient with tools such as Spectral Kurtosis evaluating the distribution of different frequency components [9] or more recently extreme value theory evaluating the distribution of spectrum excesses [10].

Nevertheless, the expensive cost of vibration sensors (such as piezoelectric accelerometers) makes these solutions often difficult to implement. Consequently, several studies [11], [12], [13] have successfully suggested to process the stator current signals. One of the advantages is that stator current measurements are often available in electrical machines for control purposes. In the same way as vibration signals, specific signatures linked to bearing faults appear on the stator current spectrum. A general review of the different techniques based on the stator current spectrum analysis is presented in [14]. These techniques are mainly based on the analysis of the stator current spectral content assessed by time–frequency [13], [15] or time-scale techniques [16]. Recent works have shown that statistical approaches such as Spectral Kurtosis [6], [17] were very efficient to monitor machine faults. These detection techniques mainly concern the induction machine. However, a few work deals with bearing fault detection in permanent magnet machines [16], [17].

Energy-based methods have been proved to be efficient in order to dissociate healthy and faulty cases afterwards but they are often very weak when it comes to predict which threshold value to use. Statistical approaches have proven to be efficient to overcome this issue as statistic-based indicators can be normalized [10], [18], [19]. More details on the computation details of the main bearing diagnostic approaches are given in [20]. Moreover, most of these works only deal with one type of machine at a certain speed and the question of reproducibility (for different speeds, for different machine of a same type) is very rarely addressed. From this point of view, the statistical approach presented in [18] shows encouraging results using center-reduced variables.

This paper focuses on bearing fault detection for a high speed permanent magnet synchronous machine PMSM belonging to an air conditioning fan used in aeronautic. Classical stator current signatures related to the vibration bearing frequencies are not suitable for such applications due to their low level. However, some frequencies multiple of the rotation frequency in the stator current spectrum have been proved to be sensitive to the considered faults [17]. Here, an indicator based on center-reduced spectral components is proposed. The interest of such a methodology is to normalize the different frequency components in order to detect and evaluate significant modifications of the spectrum without any a priori knowledge on healthy and faulty levels for the different spectral components. This technique is tested on 2 different experimental campaigns, for 4 different speeds.

The outline of the paper is the following. Section 2 starts with a description of the experimental test bench and protocol, followed by a description of theoretical signatures of bearing defects on the current spectrum. The traditional energetic indicator as well as the proposed statistic-based indicator are then detailed. The evaluation protocol is presented at the beginning of Section 3. The optimization of the proposed method as well as the results obtained with energy-based and statistic-based indicators are presented in this section. These results are then compared and discussed in Section 4.

Section snippets

System description and experimental protocol

The studied system is an air conditioning fan (Technofan LP2). It is used in most of the commercial aircrafts to provide air conditioning and renewing. The whole fan is showed in Fig. 1a. Its power is 5 kVA and its maximum speed Vmax is 14,100 rpm (rotations per minute).

This machine consists of a high-speed permanent magnet synchronous machine with sinusoidal back electromotive forces and fed by a pulse width modulation (PWM) current source inverter operating sequentially to provide 120° square

Performance evaluation metrics

The performance of the energy-based and the statistic-based indicators is evaluated on data from campaigns 1 and 2. Campaign 1 is used as a training dataset in order to optimize the different parameters of the methods. These optimized criteria are then evaluated on campaign 2 to test if they are reproducible with different bearings. The vibration indicators recorded during both campaigns are used as references to compare the performance of the different methods because they are traditionally

Discussion

Bearing faults are detected at each speed for both campaign either by energy-based or statistic-based indicators. The comparison between the results obtained with the energy-based indicator (Section 3.3) and with the statistic-based indicator (Section 3.4) shows that the use of the statistic-based indicator leads to a more accurate detection of bearing faults. Table 8 resumes the overall results on campaigns 1 and 2 for both indicators. The results presented in Table 8 are the average of the

Conclusion

An original method for bearing fault detection in permanent magnet synchronous machine is designed and evaluated in this paper. This method statistically processes the stator current spectrum in order to detect changes from a reference frequency content. The stator current spectrum is computed according to Welch׳s method in order to minimize its variance. A reference is then built on a set of spectra by calculating the mean and the standard variation for each frequency bin. As this reference is

Acknowledgments

This work is part of the French national project PREMEP [21] (PRojEt Moteur Electronique de Pilotage), labeled by the Aerospace Valley cluster and involving the LAPLACE and Airbus suppliers such as Technofan, Liebherr Aerospace, CIRTEM, DELTY and ADN. It was founded by the Fond Unique d׳Investissement, and the Aquitaine and Midi Pyrénées regions.

References (27)

  • J. Antoni et al.

    The spectral kurtosisapplication to the vibratory surveillance and diagnostics of rotating machines

    Mech. Syst. Signal Process.

    (2006)
  • R. Randall et al.

    Rolling element bearing diagnostics—a tutorial

    Mech. Syst. Signal Process.

    (2011)
  • P.J. Tavner

    Review of condition monitoring of rotating electrical machines

    IET Electr. Power Appl.

    (2008)
  • S. Jeevanand, B. Singh, B. Panigrahi, V. Negi, State of art on condition monitoring of induction motors, in:...
  • S. McInerny et al.

    Basic vibration signal processing for bearing fault detection

    IEEE Trans. Educ.

    (2003)
  • A. Sadoughi, M. Ebrahimi, E. Razaei, A new approach for induction motor broken bar diagnosis by using vibration...
  • W. Zhaoxia, L. Fen, Y. Shujuan, W. Bin, Motor fault diagnosis based on the vibration signal testing and analysis, in:...
  • J. Chebil et al.

    Wavelet decomposition for the detection and diagnosis of faults in rolling element bearings

    Jordan J. Mech. Ind. Eng.

    (2009)
  • S. Ai, H. Li, Y. Zhang, Condition monitoring for bearing using envelope spectrum of EEMD, in: Proceedings of the...
  • V. Vrabie, P. Granjon, C.-S. Maroni, B. Leprettre, Application of spectral kurtosis to bearing fault detection in...
  • A. Hazan, M. Verleysen, M. Cottrell, J. Lacaille, Probabilistic outlier detection in vibration spectra with small...
  • H. Henao et al.

    Analytical approach of the stator current frequency harmonics computation for detection of induction machine rotor faults

    IEEE Trans. Ind. Appl.

    (2005)
  • M. Blodt et al.

    Mechanical load fault detection in induction motors by stator current time-frequency analysis

    IEEE Trans. Ind. Appl.

    (2006)
  • Cited by (45)

    • A coarse-to-fine decomposing strategy of VMD for extraction of weak repetitive transients in fault diagnosis of rotating machines

      2019, Mechanical Systems and Signal Processing
      Citation Excerpt :

      Specifically, the repetitive transients caused by a weak incipient defect are prone to be overwhelmed by heavy noise, abnormal shocks, strong low-frequency vibration components, gear meshing frequency components, etc., which make the extraction of the repetitive transients become a great challenge. Advanced signal processing techniques are commonly used tools to analyze the collected raw signals to reveal the fault characteristics in fault diagnosis of rotating machines, including time series analysis [4,5], frequency transform [6], envelope demodulation [7,8], time-frequency analysis [9–11], and so on. Due to no transform basis incorporated and adaptive filtering merit, the adaptive signal decomposition methods have been extensively studied and widely applied in fault diagnosis of rotating machines, such as empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD) [2], local mean decomposition (LMD) [12], intrinsic time-scale decomposition (ITD) [13], etc.

    • Evaluating transient performance of servo mechanisms by analysing stator current of PMSM

      2018, Mechanical Systems and Signal Processing
      Citation Excerpt :

      Li et al. developed a method for monitoring tool flute breakage during end milling using the feed-motor current signals [4]. Picot et al. extracted a statistic-based indicator from the current spectrum for detecting the bearing faults of an air conditioning fan [5]. Verucchi et al. identified the misalignment between the motor and load by using the frequency components of the stator current and the estimated electrical torque [6].

    View all citing articles on Scopus
    View full text