N.S. Jammu et al.

/ International Journal of Engineering Science and Technology (IJEST)

A Review on Prognosis of Rolling Element Bearings

N.S. Jammu1 and P.K. Kankar2
John Deere India Private Limited Pune 412208, India, Email: navrajammu@gmail.com 2 Mechanical Engineering Department PDPM Indian Institute of Information and Technology, Design and Manufacturing, Jabalpur- 482005, India, Email: kankar@iiitdmj.ac.in Abstract- Bearings are amongst the frequently encountered components to be found in rotating machinery. Though inexpensive, their failure can interrupt the production in a plant causing unscheduled downtime and production losses. So the bearing prognosis plays a significant role in reducing plant down time and enhanced operation safety, by estimating the Remaining Useful Life (RUL) of damaged bearing. Admitting the importance of bearing prognosis, this literature review attempts to summarize various techniques, methods and models used in the prognosis of bearing till date. The definition of bearing prognosis is discussed in the beginning, followed by classification of various prognostic methods, review of methods used by various investigators in this research domain and concluding the topic with the summary of future research directions. Keywords- Bearing Prognosis, Remaining Useful Life (RUL). 1. INTRODUCTION
1

A bearing is a mechanical element that not only supports the load but also permits relative motion between two parts, such as shaft and the housing, with minimum friction. They are the most common components to be found in rotational machines and play an important role in the proper operation of these machineries. The presence of faults in bearings results in severe vibrations of rotating machinery. Timely detection of these faults and estimation of the time to failure are the areas of concern for researchers because abrupt failure of bearings may cause malfunctioning of the entire system and this result in downtime for the system and economic loss to the customer. A bearing failure also has the potential to damage machinery causing soaring machinery repair and/or replacement costs. Thus for the last decades, condition based maintenance (CBM) has been the subject of extensive research. Fault detection, diagnosis and prognosis are the three mainstays of CBM. Detection comprises of determining that the damage has occurred to the bearing, while diagnosis is a determination of the location and type of fault, whereas prognosis involves estimation of the remaining life of the damaged bearing and investigation of failure modes. A significant objective of CBM is to predict the machine health or the remaining useful life (RUL) instead of its service time, which leads to anticipated usage of the machine, reduction in downtime and enhanced operational safety. Assessment of bearing performance degradation is more effective than fault diagnosis to realize CBM. Palmgren and Lundberg [1-3] have given foundation of developing life prediction methods for ball and roller bearings which resulted in standards for the load ratings and life of rolling-element bearings [4-6]. An effective prognostics program provides sufficient time to schedule a repair and to acquire ancillary components before catastrophic failures occur. This literature review is organized as follows: section 2 discusses the meaning of bearing prognosis and classification of various prognosis methods. Section 3 reviews the various techniques, methods and models used in the prognosis of bearing till date. Section 4 concludes the literature review with summary and future research directions. 2. DEFINITION OF BEARING PROGNOSIS

According to the International Standard Organization [7], failure prognostic corresponds to the estimation of the time to failure and the risk for one or more existing and future failure modes. Its the art or act of predicting future conditions on the basis of present signs and symptoms [8]. The implementation of prognosis system results in an optimal maintenance schedule. Bearing prognostic activity aims at anticipating the failure date by predicting the future health state of a bearing system and

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7497

N.S. Jammu et al. / International Journal of Engineering Science and Technology (IJEST)

its RUL [9]. Most of the research related to CBM is mainly focused on fault diagnostic. Failure prognostic is a new research domain and there is a wide scope to develop methods, tools and applications for effective prognostic system. Palmgren [4] has developed functions for the service life of bearings under purely radial load and established rules for the conversion of axial and simultaneously effective axial and radial loads into purely radial loads. He also calculated the effect of different types of loads that are subject to change over time. Failure prognostic methods can be classified into three main approaches [9], as shown in the Fig. 1.

Model Based Prognostic Methods Experience Based Data Driven

Fig. 1 Classification of Prognostic Methods

A.

Model-Based Prognostic Methods

These methods make use of an analytical model (set of differential or algebraic equations) to represent systems dynamic behaviour and degradation phenomenon. Yu and Harris [10] have proposed a stress based fatigue life model for ball bearings. Its benefit lies in the fact that, it delivers more accurate prognostic results. But in actual, real systems are often nonlinear and the degradation mechanisms are generally stochastic, so difficult to obtain in the form of analytical models. Consequently, the applicability of this approach may be limited in practice. B. Experience-Based Prognostic Methods

In these methods, the data of an experience feedback collected over a significant period of time (failure times, maintenance and operating data, etc.) is used to predict the time to failure or the RUL. Their main advantage is that they are based on the use of simple reliability functions (Exponential law, Weibull law, etc.) rather than complex mathematical models. Palmgren [4] has established empirical relation between the service life of bearings and the load on bearing which is based on perception, experimental observation and curve fitting. The prognostic results from these methods are less precise than those provided by model-based and data-driven approaches, especially when the operating conditions are variable or in the case of new systems because of a lack of experience data. C. Data-Driven Methods

In these methods an online data is captured with the help of sensors and converted into relevant information. It is then used to study the degradation phenomenon based on different models and tools like (neuronal networks, Bayesian networks, Markovian processes, etc.) or statistical methods, to learn the degradation model and to predict the future health state and the corresponding RUL of the system. Shao and Nezu [11] have proposed progression-based prediction of remaining life (PPRL) of bearing. The basic concept behind PPRL is to apply different prediction methods to different bearing running stages. The procedure consists of on-line modelling of the bearing running state via neural networks and logic rules. The proposed procedure can automatically adapt to changes in environmental factors. Gebraeel et al. [12] have developed neural-network-based models for predicting bearing failures and concluded that the weighted average of the exponential parameters gives best estimate of bearing failure times. Data-Driven Methods have advantage over both of the above methods i.e. model-based prognostic methods and experience-based prognostic methods as in one hand, in real industrial applications getting

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7498

N.S. Jammu et al. / International Journal of Engineering Science and Technology (IJEST)

reliable data is easier than constructing physical or analytical behavior models. And in the other hand, the generated behavioral models from real monitoring data lead to more precise prognostic results than those obtained from experience feedback data. 3. REVIEW OF BEARING PROGNOSIS

Though failure prognostic is a new research domain, but not a new concept and has been addressed in number of publications over number of years. There are various types of bearing condition monitoring and prognosis techniques available as shown in Table 1.
TABLE 1 Summary of Typical Bearing Condition Monitoring and Prognosis Techniques

Condition Monitoring Techniques Vibration Oil/Debris Analysis Temperature Acoustic Emission

Advantages Reliable and standardized method Reacts immediately to change Ability to point out defective component Disadvantages Expensive Intrusive Subject to sensor failures

Advantages Ease of bearing condition characterization Possible to detect type and location of defect Disadvantages Expensive Limited to bearings having oil supply system (difficult to apply on grease lubricated bearings)

Advantages Ability to detect anomalous working areas Standardized method Disadvantages Not efficient since substantial rise in temperature occurs in last stages of life Difficult to detect impending failure

Advantages Impending fault can be detected Large signal-tonoise ratio Disadvantages Expensive Sampling frequency must be high

Earlier bearing prognosis made use of vibration analysis, in which the current and previous vibration data was used to predict the RUL of a bearing. Khan [13] and Succi [14] used a simple test rig bearing failure mode for bearing prognosis. Numbers of endurance tests are conducted on bearings of test rig with damages on the outer race, inner race and rolling element of the bearing and was run between 100 to 900 hours before failure. Khan [13] has also displayed double log of the Resonance Response Measurement (RMS) values i.e. ln(ln(RMS)), against a linear time scale. He found that a straight line was obtained from which a least square fit could be used to estimate the safe operating life with a definable confidence band. A number of times domain parameters had been computed and trended throughout the various tests including RMS, peak, crest factor, Kurtosis, ratio of peak in 5-12.5 kHz to 0-5 kHz frequency band, ratio of RMS in 5-12.5 kHz to 0-5 kHz frequency band and transducer resonance response measurement (RMS) in 27-35 kHz frequency band [15]. Frequency band of 5-12.5 kHz was found to provide the most sensitive results out of the number of frequency bands used for the time domain statics due to the presence of number of structural resonances. In actual practice the vibration environment and range of failure modes vary considerably so it is difficult to extrapolate the results to general industrial rotating machinery. Randall [16] and Berry [17] have focused on prognosis of bearing in general industrial rotating machinery and contributed to overcome limitation of model proposed by Khan. Randall recommended the use of fitting a straight line to the db values (logarithmic scale) against a linear time scale. The curve

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7499

N.S. Jammu et al. / International Journal of Engineering Science and Technology (IJEST)

drawn was extrapolated to reach 20 db change with respect to the baseline level, to obtain the predicted lead time. However, a number of faults like spall and crack development show a sudden stepwise trend change, which makes it difficult to predict the trend with this method. In the area of bearing prognosis, the work by all of the above laid the foundation of significant progress to be made. However, most of the previous studies have not considered the effects of changes in operational environment and machine running conditions like change in load, speed and equipment repairs. The procedure of predicting bearing life should have adaptive prognostic ability to provide highly accurate automatic long term prediction. This was overcome by prediction procedure referred to as, Progression Based Prediction of Remaining Life (PPRL) [11]. It utilizes various variables characterizing the state of a degradation mechanism determined from on-line measurements. The applications of PPRL include forecasting a bearing defect development process, the exact RUL of bearing and to solve problems of adaptive environmental factor changes. Li et al. [18] have also attempted to overcome the limitations of condition monitoring techniques such as damage severity estimation and prognostic capability. The concept of a defect propagation model in conjunction with adaptive prognostic methodology for the prediction of bearing defect growth was introduced. Comparative study had been carried out between future bearing defect size as a forecast by prognostic model and measurement inferred bearing condition. It showed certain amount of prediction error owing to the stochastic nature of defect growth. Fine tuning of the model parameters had been done using adaptive algorithm. They concluded that the adaptive prognostics effectively predict the bearing defect propagation process without the need for previous knowledge of prognostic model parameters. Indirect or non intrusive measurements, such as vibration and acoustic emission can be effectively used to estimate defect area size through signal processing and diagnostic models without any need to stop the machinery operations. Another work carried out by Orsagh et al. [19] consists of both bearing spall initiation and progression models used in aggregation with the feature analysis critical engine bearings to predict the RUL of critical engine bearings. Model based approach provides the valuable damage build-up information on critical components well in advance of failure indications, while direct measures of component condition that can be used to update the modelling assumptions and reduce the uncertainty in the RUL predictions, is provided by sensor based approach. Thus the models together offer complementary condition assessment information that can be used to achieve a comprehensive prognostic capability throughout a components life. The signature of faulty bearings is often masked by noise and low frequency effects. For accurate and precise prognosis of bearing system the signature of a defective bearing should be clear, so measuring mechanism should be efficient and capable of measuring signature across a wide frequency band. Qiu et al. [20] have introduced enhanced and robust prognostic methods for rolling bearing. These methods include a wavelet filter based approach for weak signature enhancement for fault identification and Self Organizing Map (SOM) based method for performance degradation assessment. To validate the methodology a rolling bearing run-to-failure test was performed, which was designed to generate multiple sets of full life cycle degradation data. The results demonstrated that the bearing defects can be detected at an early stage of development by designing an optimal wavelet filter. The results when combined with time series prediction methods such as Auto Regressive and Moving Average (ARMA) [21] and recurrent neural network [22], the prediction of remaining useful life of rolling bearing can be realized [19]. Mandic and Chambers [22] studied bearing spall progression and used a physics-based model. The two key elements are identified to model spall propagation. First is determination of dynamic loads and stresses (using numerical approaches) occurring in the material as a rolling element passes over the spall and the other is development of a method relating this local stress field to damage accrued in the material. Based on stress information, damage mechanics algorithm is used to determine spall propagation rate. Reduced Order Model (ROM) is used to calculate material removal rate and diagnostic feedback is used to self-adjust and reduce prediction uncertainty. The results were verified on both subscale bearings and full-scale turbine engine bearings and demonstrated that spall propagation can be predicted with high confidence. Being physics based, this model is readily adaptable to most bearing applications. Kankar et al. [23] presented a methodology for rolling element bearings fault diagnosis using continuous wavelet transform (CWT). Three machine learning techniques have been used for faults

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7500

N.S. Jammu et al. / International Journal of Engineering Science and Technology (IJEST)

classications, out of which two are supervised machine learning techniques, i.e. support vector machine (SVM), articial neural network (ANN) and other one is an unsupervised machine learning technique, i.e. self-organizing maps (SOM). In their further work, Kankar et al. [24] proposed a fault diagnosis methodology for rolling element bearings, which utilizes autocorrelation of raw vibration signals to reduce the dimension of vibration signals with minimal loss of significant frequency content. It is observed that dimension of vibration signal is reduced to 10% with negligible loss of information. Goebel et al. [25] have suggested the fusion of independent estimates from two fundamentally different methods to estimate remaining life. First method uses Physics Based (PB) model to forecast future bearing condition and returns a probability density function of the bearing remaining useful life with the help of historic and estimated future operating conditions. In this method, principle of damage mechanics are used to track material microstructure changes, so takes into account material properties, bearing geometry, surface interaction, lubrication, and variable operating conditions. The other method uses an empirical model to estimate condition-based fault propagation rate, using data from experiments at known conditions and component damage level. These two approaches are fused on the basis of Dempster-Shafer regression to produce a result that is more accurate and more robust than either method alone. Experimental results from rig tests, in which spall was initiated on a bearing and its growth carefully monitored, are nearly the same as that of prognostic estimates of the reasoner. Janjarasjitt et al. [26] have used a modied form of the correlation integral developed by Grassberger and Procaccia. Machine vibration data obtained throughout a life test of a rolling element bearing is then analysed by using the partial correlation integral algorithm. Results showed that the dimensional exponent as computed from the partial correlation integral algorithm tends to increase as time progresses and the useful remaining life of the bearing is decreasing. Thus if a proper threshold of the dimensional exponent is chosen, it can be used to predict the RUL of bearing. But the methodology need to be further validated through more experiments, modelling and computational work. Ghafari [27] in his doctoral work employed vibration signals and three monitoring indices namely Normalized Information Entropy (NIE), Correlation dimension (CD), and Maximum Approximate Coefficient of Wavelet (MACW), as the input of the diagnostic scheme. The output of the system corresponds to the level of bearings health. Recurrent Neural Networks (RNN) and Adaptive NeuroFuzzy Inference System (ANFIS) are used to evaluate the efficiency of bearing prognosis. It was concluded that once an ANFIS system is trained with the run-to-damage vibration data of a bearing, the network can capture the damage propagation behavior accurately and can be utilized successfully to predict future states of the same series of the bearings under different speed and load conditions. The test results confirm a success rate greater than 98 percent. For accurate assessment of the residual life of bearings, Kim et al. [28] proposed a machine prognostics model based on health state estimation using Support Vector Machines (SVM). Bearing fault cases of High Pressure Liquefied Natural Gas (LNG) pumps were analysed to obtain the failure degradation process of bearing failure and to validate the feasibility of the proposed model. Kankar et al. [23, 29] have also shown the effectiveness of SVM for bearing faults classification. However, effective feature extraction techniques for a variety of faults are still needed. In addition to above models, a hybrid model has been proposed [30], to assess bearing performance degradation by combining benefits of Support Vector Data Description (SVDD), with excellent computation and Fuzzy C-Means (FCM), an algorithm with excellent explicable results. The clustering centre of normal state is obtained by SVDD. The subjection of tested data to normal state, which is computed using FCM algorithm with final failure data, defined as a degradation indicator. Accelerated bearing life test was performed to check the effectiveness and robustness of proposed hybrid model. In a one of the recent investigation [9], the use of Mixture of Gaussians Hidden Markov Model (MoGHMM) was proposed for prognosis of bearings. Extracted features are used to learn different models that capture the time evolution of degradation and therefore of the systems health state in an off-line step. While in on-line step, failure diagnostic and prognostic are done by estimating the assets current health state and its RUL from learned models. But future works are needed to compare the proposed method with existing ones using the same data in order to assess its pros and cons. In addition, it would be interesting to test the method on a real industrial system.

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7501

N.S. Jammu et al. / International Journal of Engineering Science and Technology (IJEST)

4.

CONCLUSIONS

This paper has briefly reviewed state of art of bearing condition monitoring and prognosis. The emphasis has been put on various prognosis methods used for bearings. Although lot of work has been carried out in the area of bearing fault diagnosis but area of bearing prognostics need to be explored to estimate correctly RUL of bearing. Further development is required particularly for difficult cases of rotating machinery in practical applications, where number of failure modes could be expected due to different loading and operating conditions. Accurate bearing prognosis requires signals from several sensors which are costly and difficult to install in already mounted industrial machines. Also number of methodologies discussed above needs further study, in order to verify their application to industrial problems with varied experimental conditions. The gaps found in the area bearing prognosis are as follows: 1) Signals received by industrial machines are often complex, non-stationary and noisy nature. Such signals require large number of sensors. Optimal quantity and location of these sensors need to be identified because sensors failures and their locations can also reduce reliability of prognosis system. 2) More variety of faults need to be studied and modelled for accurate bearing prognosis since its accuracy is subjected to stochastic processes that are yet to occur. 3) The interaction effects of faults need to be considered. 4) Variable operating conditions and the stochastic characteristics of signals prevent the usage of traditional time or frequency domain techniques. Time-frequency analysis such as wavelet transforms need to be explored because it can be effectively used in such situations. 5.
[1] [2] [3] [4] [5] [6] [7] [8] [9]

REFERENCES

[10] [11] [12] [13] [14] [15] [16] [17] [18]

[19] [20] [21]

Palmgren, Ball and Roller Bearing Engineering, First edition, Translation by G. Palmgren and B. Ruley, SKF Industries, Inc., Philadelphia, PA, 1945. G. Lundberg and A. Paimgren, Dynamic Capacity of Rolling Bearings, Acta Polytechnica Mechanical Engineering Series, vol. 1, no. 3, Stockholm, Sweden, 1947. G. Lundberg and A. Palmgren, Dynamic Capacity of Rolling Bearings," Acta Polytechnica Mechanical Engineering Series, vol. 2, no. 4, Stockholm, Sweden, 1952. Palmgren, Die Lebensdauer von Kugellagern (The Service Life of Ball Bearings), Zeitschrift des Vereines Deutscher lngenieure, vol. 68, no. 14, pp. 339-341,1924,. Anon, Load Ratings and Fatigue Life for Ball Bearings, ANSI/AFBMA 9-1990, The Anti-Friction Bearing Manufacturers Association, Washington, DC, 1990. Anon, Rolling Bearings-Dynamic Load Ratios and Rating Life, ISO 281:1990(E), International Organization for Standardization, 1990. AFNOR, Condition monitoring and diagnostics of machines prognostics - part 1: General guidelines. NF ISO 13381-1, 2005. C.P. Henry, An historical view of the MFPT Society. In Proceedings of a Joint MFPTJOAPTSC Conference, Mobile, Alabama, pp. 315, 2226 April 1996. D.A. Tobon-Mejia, K. Medjaher, N. Zerhouni and G. Tripot, A Mixture of Gaussians Hidden Markov Model for failure diagnostic and prognostic, 6th Annual IEEE Conference on Automation Science and Engineering (Case), Toronto, Canada, pp. 338-343, 2010. W.K. Yu and T.A. Harris, A new stress-based fatigue life model for ball bearings, Tribology Transactions, vol. 44, no. 1, pp. 11-18, 2001. Y. Shao and K. Nezu, Prognosis of remaining bearing life using neural networks, Proc. Inst. Mech. Eng., J. Syst. Control Eng., pt. 1, vol. 214, no. 3, pp. 217230, 2000. N. Gebraeel, M. Lawley, R. Liu and V. Parmeshwaran, Residual Life Predictions From Vibration-Based Degradation Signals: A Neural Network Approach, IEEE Transactions On Industrial Electronics, vol. 51, no. 3, pp. 694-700, 2004. A.F. Khan, Condition monitoring of rolling element bearings: A comparative study of vibration based techniques, Ph.D. Thesis, University of Nottingham, May 1991. G.P Succi, Prognostic methods for bearing condition monitoring, Proceedings of the 3rd International Machinery Monitoring and Diagnostic Conference, Las Vegas, Nevada, December, pp. 335-342, 1991. Howard, A Review of rolling Element Bearing Vibration Detection, Diagnosis and Prognosis, DTSO Research Report 0013, October 1994. R.B. Randall, Computer aided vibration spectrum trend analysis for condition monitoring, Maintenance Management International, vol. 5, pp 161-167, 1985. J.E. Berry, How to track rolling element bearing health with vibration signature analysis, Sound and Vibration, pp 24-35, November 1991. Y. Li, C. Zhang, T.R. Kurfess, S. Danyluk and S.Y. Liang, Diagnostics and prognostics of a single surface defect on roller bearings, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 214, pp. 1173-1185, 2000. R. Orsagh, M. Roemer, J. Sheldon and C.J. Klenke, Prognostics/ Diagnostics for gas turbine engine bearings, Proceedings of IGTI TurboExpo 2004 Vienna, June 14-17, 2004. H. Qiu, J. Lee, J. Lin and G. Yu, Robust performance degradation assessment methods for enhanced rolling element bearings prognostics, Advanced Engineering Informatics, vol. 17, Issues 3-4, pp. 127-140, 2003. G.E. Box, G.M. Jenkins, Time Series Analysis: forecasting and control, San Francisco, CA: Holden-Day, 1976.

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7502

N.S. Jammu et al. / International Journal of Engineering Science and Technology (IJEST)
[22] D.P. Mandic and J.A. Chambers, Recurrent Neural Networks for Prediction: learning algorithms, architectures and stability,Wiley, New York, 2001. [23] P.K. Kankar, Satish C. Sharma and S.P. Harsha, Fault Diagnosis of Ball Bearings Using Continuous Wavelet Transform, Applied Soft Computing, vol. 11, pp. 2300-2312, 2011. [24] P.K. Kankar, Satish C. Sharma and S.P. Harsha, Rolling Element Bearing Fault Diagnosis using Autocorrelation and Continuous Wavelet Transform, Journal of Vibration and Control, Accepted, doi:10.1177/ 1077546310395970, 2011. [25] K. Goebel, N. Eklund and P. Bonanni, Fusing Competing Prediction Algorithms for Prognostics, IEEE Aerospace Conference, March 2006. [26] S. Janjarasjitt, H. Ocak and K.A. Loparo, Bearing condition diagnosis and prognosis using applied nonlinear dynamical analysis of machine vibration signal, Journal of Sound and Vibration, vol. 317, pp. 112126, 2008. [27] S.H. Ghafari, A Fault Diagnosis System for Rotary Machinery Supported by Rolling Element Bearings, A Thesis presented to the University of Waterloo, Ontario, Canada, 2007. [28] H.E. Kim, Andy C.C. Tan, J. Mathew, Eric Y. H. Kim and B.K. Choi, Machine prognostics based on health state estimation using SVM, Proceedings Third World Congress on Engineering Asset Management and Intelligent Maintenance Systems Conference, vol. 199, pp. 834-845, 2008. [29] P.K. Kankar, Satish C. Sharma and S.P. Harsha, Fault Diagnosis of Ball Bearings Using Machine Learning Methods, Expert Systems with Applications, vol. 38, no. 3, pp. 1876-1886, 2011. [30] Y.N. Pan, J. Chen and G.M. Dong, A hybrid model for bearing performance degradation assessment based on support vector data description and fuzzy c-means, Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, vol. 223, pp. 2687-2695, 2009.

ISSN : 0975-5462

Vol. 3 No.10 October 2011

7503