Assessing the Condition and Lifetime of Rolling Element Bearings from a Single Measurement

By:

Dr. Alexei V. Barkov, Chairman and Dr. Natalia A. Barkova, Scientific Director, VibroAcoustical Systems and Technologies (VAST), Inc., St. Petersburg, Russia,

and

John S. Mitchell, San Juan Capistrano, California

ABSTRACT

The method of rolling element bearing diagnostics using spectral analysis of the high frequency vibration envelope began in the middle 1970’s. At first the vibration analysis was directed to bearing excitation generated mainly by shock pulses. Later, random vibration excited by friction forces was included in the analysis. The method significantly improved results. By the middle 1980’s new algorithms for diagnostics and condition prediction of the rolling surfaces were developed. These algorithms enabled defining the type and depth of multiple, simultaneous defects from a single vibration measurement. By 1990 the diagnostic algorithms were automated and automatic diagnostic systems for rolling element bearings began to be used within many enterprises. A large amount of statistical data has been accumulated while watching defect development in rolling element bearings. The method, use and results of statistical data analyses are discussed in this paper.

INTRODUCTION

The single measurement method for bearing diagnostics and condition assessment is effective when these parameters are dependent solely on the bearing characteristics and detected defects but not on machine construction or the mode of machine operation. A number of friction force parameters, excited by high frequency random vibration of the bearing, fully meet these requirements. In a bearing that has no defects, the friction forces and vibration do not depend on the turning angles of the rolling elements. However, within a faulty bearing such dependence appears, namely, amplitudes periodically change. The period (frequency) of such changes defines the type of defect. The modulation amplitude defines severity.

METHOD

Identifying a rolling element bearing defect (type of defect and its severity) begins with a spectrum analysis of an enveloped random vibration signal in the range from 2 kHz to 10 kHz within a frequency band from 20% to 100% of the center frequency. An example of an enveloped vibration spectrum from a bearing with outer race wear is displayed in figure 1. When there are no defects in the bearing there are no harmonic components in the envelope spectrum. When a defect appears, its severity is defined by the difference between the level of the maximum component and the level of the random background according to the expression:

where:

mi is the amplitude of modulation of the vibration that defines the depth of the defect;

is the difference in the levels of the harmonic component and the background;

is the width of one channel in the spectrum analyzer

is the width of band pass filter used to extract the frequency band where the enveloping is made.

Within the bearing, friction forces are accompanied by periodic shock pulses. The shock pulses also become the source of amplitude modulation of the random vibration signal. Shock pulses are detected in the envelope spectrum by the appearance of a number of harmonic components with about the same amplitudes. So, by analyzing the enveloped vibration spectrum it is possible to detect and define the cause of the modulated friction forces and shock pulses. This includes all the main bearing defects that appear during installation and operation. Moreover, the method requires only one measurement with very few restrictions on the location and method of the vibration transducer mounting.

ADVANTAGES

There are three main advantages of the single measurement method compared to others frequently used for rolling element bearing diagnostics. First is the high sensitivity of the method. Vibration modulation produced by a small defect with a depth of about 1% is easily detected. Severe defects are characterized by depths of 10% and greater. The second is the ability to gain an accurate assessment of condition from a single vibration measurement in a comparatively short period of time (about 100 shaft revolutions). The third advantage is that the requirements for making the measurements are not very strict. In the simplest case it is necessary to measure vibration in the frequency range from 2 kHz to 10 kHz without the necessity for comparison with the results of previous measurements.

APPLICATION

For reliable rolling element bearing condition assessment and lifetime prediction it is necessary to have information about the type, severity and the rate of development of all the defects that have occurred due to bearing installation and operation. The single measurement method enables identification of installation defects such as increased radial tension, race misalignment and race rotation in the mounting. It also detects operating defects such as wear of the rolling surfaces or appearance of spalls (cavities or cracks) on bearing surfaces. The diagnostic symptoms of all these defects are known [1,2]. Lubrication defects are identified by an increase of the high frequency vibration from the bearing, either by a comparison with identical bearings in similar machines, or by comparison with the results of the previous vibration measurements recorded on the diagnosed bearing.

Besides the possibility of detecting and identifying all defects that limit a bearing’s operating life, long term condition prediction requires an assessment of the maximum rate of defect development. The main way to get such an assessment is to watch the development of the defects from the moment of their detection up to the moment of bearing failure. Nowadays there are data of these observations for more than a thousand bearings in different machines, operating in adverse conditions in ships, electric power plants, paper and pulp mills, metallurgical plants and in a variety of engineering plants. The analysis of these data shows that defects of rolling surfaces develop very unevenly, converting from one type into another. Moreover, there can be cases during operation when the severity of the defect, indicated by vibration characteristics, decreases. Sometimes indications of the defect even disappear and do not appear again over a large time period comparable with the mean time before failure (MTBF).

Figure 2 illustrates a frequently encountered function of the type and severity of rolling element bearing defects against operating time expressed in MTBF. The example shown in figure 2 is a hard loaded bearing in a machine with a massive horizontal rotor. Figure 2 shows that the first stage of wear, a defect in the bearing’s outer race, developed quickly (at first wear appeared, then a cavity). The defect reached the maximum severity relatively quickly and then stabilized for some time. Then the severity decreased (burnishing of the rolling surface). In time, comparable with the MTBF, another group of defects appeared in the bearing. Among these defects a rolling element flaw developed most quickly.

Figure 2 below.

Two conclusions can be generalized from this type of data obtained from many bearings:

  1. Defects begin to develop quickly in a bearing when flaws exist simultaneously in multiple rolling surfaces and;
  2. When there is defective lubrication, both conditions define the rate of wear development on rolling surfaces.

Monitoring defect development with vibration parameters also has shown that shortly before failure the increase of harmonic amplitudes in the vibration spectrum and its envelope ceases on many bearings with defects. In some cases the defect symptoms even disappear from the spectrum. However, at that time vibration increases in wide bands of frequencies. This phenomenon is caused by the increase in the number of defects on each of the rolling surfaces and their combined influence on the characteristics of the oscillating forces that operate in the bearing. The periodicity of these forces is disturbed because of the deep amplitude and frequency modulation, which cause a widening of the components in the spectrum and its envelope. It follows that to define the condition of bearings just before their failure due to multiple defects it is impossible to use spectral methods of diagnostics. Under these conditions envelope spectrum assessment must be augmented by wide band vibration energy measurements. In this final stage of rolling element bearing service life, characterized by quick degradation, it is possible to make an assessment of a bearing’s remaining operating life. The traditional methods of trend analysis can be used in these cases. The trends are formed by the results of periodic vibration measurements of vibration characterizing the increase of energy components in wide frequency bands.

RESULTS

Observation of the symptoms of defect development in a bearing is a fast, efficient method of condition assessment, but it does not usually provide reliable results. This occurs because the information gained by this observation is used by the customer as feedback to correct the machine operating condition or to initiate bearing replacement. To get a reliable assessment, data from single diagnostics tens of thousands of bearings were studied. About five thousand of these bearings had one defect in the incipient stage of development and were used in the study. The result of correlating failures during their remaining operating life to MTBF is shown in figure 3. In total, 1,940 bearings were detected with defects on the outer race, 1,276 bearings with defects on the inner race, 486 bearings with defects on the rolling elements and 1,292 bearings with lubrication defects. As seen from these data, wear of a bearing most often begins from the outer race and least often from the rolling elements. But, at the same time, rolling element defects develop most rapidly. During the observation time, bearing failures occurred to 25.6% of the bearings where the wear began from an outer race defect, 28.1% of the bearings where a lubrication defect was the first detected, 44.1% with an inner race defect and 76.3% where a rolling element defect was the first detected. In total, 35.8% of the controlled bearings with different defects were replaced.

Figure 3 (below)

USE

From the data it is possible to assess the time from recognizing a defect within a rolling element bearing up to the bearing’s failure. This time is not less than 25% of MTBF with a certainty of about 0.98. This leads to a conclusion that a rolling element bearing will operate for at least 20% of its MTBF before failure following an accurate condition assessment where defect symptoms are not present. This value is the maximum time period between long term condition prediction measurements that can be safely allowed by the single measurement method. Making the next diagnostic measurement within this time period and not detecting any defects provides repeated assurance of a predicted lifetime that is independent of a bearing’s operating time and the number of previously made diagnostic measurements. It should be noted that in some cases experience in condition prediction may result in giving recommendations on the interval between diagnostic analyses on bearings with an MTBF of greater than three years. The maximum time period between diagnostic measurements should be no greater than six months if no defects are detected. Such an interval is usually sufficient to minimize unforeseen bearing faults due to violation of machine operating requirements, machine overloading for example, operating outside specified environmental conditions and so on.

In the case when a defect is detected, the interval between bearing predictive condition measurements can be determined in the same way, but with reduced intervals between measurements. The detected defects can be divided into three main groups. Incipient defects usually do not influence the bearing’s condition and the results of condition prediction. Medium defects change the bearing’s condition and the interval between condition prediction measurements, but do not increase the probability of bearing failure within the interval between condition prediction measurements. Severe defects increase the probability of a failure, but in the limits defined by the customer (usually up to 0.05). When medium defects are detected, reducing the condition prediction interval to one half the normal interval is recommended. When one severe defect is detected, the interval between measurements should be reduced by a factor of five to six compared to the condition prediction interval determined for a bearing without defect symptoms.

Experience shows that the probability of a bearing failure sharply increases during the final period of a bearing’s service life (the stage of degradation) when several defects are developing simultaneously on different bearing elements and one has exceeded the severe defect level. Under these extreme conditions with symptoms of a severe defect present in the vibration signal, a recommendation can be made not to replace the bearing, provided there are no other defects. However, in such unusual and extreme cases, the diagnostic measurements must be accomplished at close intervals to assure the appearance of new defects, indicative of imminent failure, are not missed.

The certainty of bearing condition prediction and diagnostics by single envelope spectrum measurements can be increased if diagnostic data is recorded on a new bearing after a few days of operation. By conducting diagnostics shortly after installation, defects such as race misalignment and increased radial tension are detected. These lead to significant growth of the loads on the rolling surfaces and must be avoided. Although these loads decrease with bearing wear, their existence in the period of bearing run in leads to a reduction in the bearing’s resources. The main point is that faulty installation leads to a significant increase in the rate of defect development that can appear later in the bearing’s service life, even though loads have not increased. If the results of diagnostics conducted in the beginning stage of bearing operation are known, then the mounting defects can be corrected or condition measurements recorded two to three times more often than required for a non defective bearing.

Experience with periodic bearing diagnostics and long term condition prediction shows that during the operating life of a properly installed and lubricated bearing only about 10 to 20 envelope spectrum condition prediction measurements are required to provide assurance against unexpected faults. Moreover, most bearings can operate successfully for several MTBFs before replacement becomes necessary based on actual condition determined from accurate diagnostics.

SUMMARY

1. From a diagnostic perspective, the operating life of a bearing consists of four main stages - run in, operating without defects, development of defects and bearing degradation to eventual failure.

2. Spectrum analysis of a single enveloped, high frequency, random vibration signal generated by a rolling element bearing can be an effective method of condition assessment and an accurate means of lifetime prediction during the first three stages of a bearing’s operating life.

3. During the last stage, which is characterized by the simultaneous development of several defects on different rolling surfaces, methods based on a narrow band spectral analysis of the enveloped vibration signal must be supplemented by a total energy measurement.

4. Ten to twenty aaverages of the spectrum of a bearing’s enveloped high frequency random vibration signal is sufficient to minimize the chance of non-predicted faults within a properly installed and operated rolling element bearing, and to replace a bearing according to its actual condition.

REFERENCES

1. Alexandrov A.A., Barkov A.V., Barkova N.A., Shafransky V.A. -Vibration and Vibrodiagnostics of Electrical Equipment in Ships, -Sudostroenie (Shipbuilding), Leningrad, 1986.

2. Azovtsev Yu.A., Barkov A.V., Yudin I.A., “Automatic Diagnostics of Rolling Element Bearings Using Enveloping Methods,” Proceedings of the Vibration Institute 18th Annual Meeting, pp. 249-258 (1994).

Return to Inteltek home page Return to articles/notes page