Automatic bearing fault diagnosis based on one-class ν-SVM

https://doi.org/10.1016/j.cie.2012.10.013Get rights and content

Abstract

Rolling-element bearings are among the most used elements in industrial machinery, thus an early detection of a defect in these components is necessary to avoid major machine failures. Vibration analysis is a widely used condition monitoring technique for high-speed rotating machinery. Using the information contained in the vibration signals, an automatic method for bearing fault detection and diagnosis is presented in this work. Initially, a one-class ν-SVM is used to discriminate between normal and faulty conditions. In order to build a model of normal operation regime, only data extracted under normal conditions is used. Band-pass filters and Hilbert Transform are then used sequentially to obtain the envelope spectrum of the original raw signal that will finally be used to identify the location of the problem. In order to check the performance of the method, two different data sets are used: (a) real data from a laboratory test-to-failure experiment and (b) data obtained from a fault-seeded bearing test. The results showed that the method was able not only to detect the failure in an incipient stage but also to identify the location of the defect and qualitatively assess its evolution over time.

Highlights

► The aim of this work is to develop a new automatic system for monitoring the condition of rolling-element bearings. ► The method is able not only to detect the failure in an incipient stage but also to identify its location. ► The parameters of the method are adjusted using only normal functioning data following a novelty detection philosophy.

Introduction

Machine condition monitoring has emerged as a strategic area of concern in most companies since unanticipated faults in their production assets lead to losses that can affect their efficiency and productivity. Predictive maintenance is the most effective practice and, nowadays, is being implemented in many maintenance departments. It is based upon computer aided methods continuously monitoring the condition signals of a machine and periodically analysing its status.

Vibration analysis is the most popular technique in machine condition monitoring (Taylor, 2003, Scheffer and Girdhar, 2004). Existing principles that explain the signs that common failures leave behind in the raw vibration signal are well determined and thus can be used in the design of predictive maintenance systems. Many vibration phenomena can be interpreted as an amplitude modulation of the characteristic vibration frequency of the machine. Envelope analysis is a classical demodulation technique that emerged from the signal processing field already proven useful in the uncovering of early faults (Jones, 1996, McInerny and Dai, 2003, Yang et al., 2007b).

Bearing faults are among the most common problems encountered in the use of high-speed rotating machinery. The reason for this being that almost any industrial machine has at least one of these components and their fault can be the direct cause of subsequent problems in other vital components. Furthermore, because the time to catastrophic failure is different for inner race, outer race, rolling element and cage defects, it is very important to know the nature and severity of a bearing fault in order to select the most appropriate maintenance action.

Fault diagnosis in rolling-element bearings has been an important research topic in pattern recognition over the last decade. However, most studies have centred on fault type classification based on the availability of fault samples (Fang and Zijie, 2007, Yang et al., 2007a, Al-Raheem and Abdul-Karem, 2010, Kankar et al., 2010, Meng et al., 2010, Wang and Chen, 2011, Wu et al., 2012). But in real life applications it is extremely difficult to get data for all types of bearing problems because they do not occur frequently, and furthermore, failure vibration patterns are machine-specific. Some previous research have treated the problem as a novelty detection task (Mitoma et al., 2008, Pan et al., 2009, Alzghoul and Löfstrand, 2011, McBain and Timusk, 2011). Novelty detection is concerned with recognising inputs that differ in some way from those that are usual under normal conditions (Marsland, 2002). This paradigm overcomes one important limitation of competing methods in machinery fault detection problems i.e. the need for pre-collecting failure data. But these works are usually restricted to the detection of a fault.

The aim of this work is to develop a new automatic system for monitoring and diagnosing the condition of rolling-element bearings. The base of the system is an early detection of failures using a one-class ν-SVM (Schölkopf et al., 2000). This model treats fault uncovering as a novelty detection problem. Together with fault detection, an analysis of the envelope spectrum of faulty signals by means of a knowledge-based system, is proposed to diagnose whether the defect is in the inner race, outer race, rolling element or cage.

The contents of this paper are organised in the following way. In Section 2, background on rolling-element bearings is introduced. In Section 3, both the methodology proposed to detect and diagnose faults in bearings and the description of the techniques involved in the main steps of the method are presented. In Section 4, the proposed method is validated using data from two different real scenarios. Finally, the conclusions of this work are given in Section 5.

Section snippets

Rolling-element bearings

A rolling-element bearing is a mechanical device that reduces the friction between a rotating shaft and two or more pieces connected to it. The main components of a rolling bearing are: outer race, inner race, rolling elements and cage (see Fig. 1a). Each time a defect on one surface of a component strikes another surface, a force impact is produced. If the rotational speed of the races is constant, the impact repetition rates can be determined by the geometry of the bearing (McFadden and

Model description

It is a common practice in machine monitoring to extract explanatory parameters from the raw vibrational data. The most frequently used methodology is first to extract some global statistics such as the root mean square (RMS) in order to detect a deviation, and subsequently to calculate its power spectrum (usually through Fast Fourier Transform) in order to analyse predetermined sub-bands and detect the location of the fault. If there is a fault, the spectrum will change in comparison to normal

Experimental results

Two different data sets have been used to check the performance of the fault diagnosis method proposed in the previous section: one with data obtained from a laboratory test-to-failure experiment and another with data from a fault-seeded bearing test. In every experiment, accurate fault detection has been achieved using the following parameters for the ν-SVM: ν = 0.01 and a Gaussian kernel with γ = 0.05 (see Eq. (8)). The first parameter represents the estimation of spurious data (1%) in the normal

Conclusions

In this paper, a new automatic method to diagnose faults in bearings based on pattern recognition and signal processing techniques is presented. The main contribution of this work lies in the combination of ν-SVM, envelope analysis and a rule-based expert system in order to early detect and diagnose the defective component of the bearing. An important stage of the method is the proposed Sensitivity Test which is able to automatically select the zone of the vibration spectrum where the fault is

Acknowledgements

This work was supported in part by the Spanish Ministry of Science and Innovation (MICINN), Grant code TIN2009-10748, and by the Xunta de Galicia, Project CN2011/007, both partially supported by FEDER funds.

References (31)

  • Fang, S., & Zijie, W. (2007). Rolling bearing fault diagnosis based on wavelet packet and RBF neural network. In...
  • R. Fletcher
    (1987)
  • T. Harris

    Rolling bearing analysis

    (1991)
  • R. Jones

    Enveloping for bearing analysis

    Sound and Vibration

    (1996)
  • P. Kankar et al.

    Fault diagnosis of ball bearings using machine learning methods

    Expert Systems with Applications

    (2010)
  • Cited by (0)

    This manuscript was processed by Area Editor Min Xie.

    View full text