Condition Assessment and Life Prediction of Rolling Element Bearings, Part 2

Alexej Barkov, Natalja Barkova, VibroAcoustical Systems and Technologies, St.Petersburg, Russia,

edited by John S. Mitchell

Part 1 introduced the flaw characteristics of rolling element bearings, described how they can be extracted from an external vibration signal and discussed several methods for condition assessment and lifetime prediction. Part 2 picks up where part 1 ended. Actual field experience, an explanation of defect development. more on lifetime prediction and benefits demonstrated during many years of use in Russia are all covered.

Field Experience

Assessing the condition of rolling element bearings by an enveloped high frequency random vibration spectrum is regularly accomplished by many enterprises in Russia. Condition assessment intervals vary from not less then twice a year on defect free bearings to much more often following defect detection. For the past several years more than ten thousand monitored bearings were replaced based on external measurements of condition. Half of the bearings removed were examined visually. The resulting statistical data provide an accurate estimate of the maximum rates of development for all types of defects detected by enveloped vibration spectra.

Cause of Bearing Failures During Initial Operation

The principal causes of bearing failures during the beginning period of operation (about 20% of bearing service life) is the first result of this analysis. The number of bearing failures during initial operation approached 10% of all bearings that failed during the test period. This was about the mean service life of the bearings installed in the different types of machines included in the study. Approximately two thirds of the bearings that failed early in life had installation defects. Among the defects found most frequently were increased radial tension and misalignment of the fixed bearing race. Many of the failed bearings (about half) had been operating outside specified operating conditions for some period of time. There were also cases where the machine was overloaded, operated at excessive temperatures, with water or other contaminants in the lubrication system and other similar conditions. Not one beginning of life failure was experienced on properly installed bearings operating within specified conditions.

It should be noted that only a few rolling element bearings where installation defects were detected reached a failure condition during the initial period of operation. For example, installation defects of varying severity were detected in about one third of the total bearings studied. However, the number of failed bearings from all causes never exceeded one third of the total bearings in the study. At the same time, bearings with mounting defects failed much more frequently (three to five times) during the final stage of operation from defects on the inner race, specifically, wear, cavities and cracks. These same defects were found in the first bearings removed and replaced during the beginning period of operation.

Defect Development during Bearing Life

Several peculiarities were noted during the course of observing bearing failure characteristics. Non-uniform defect development was not uncommon, even after the appearance of a severe defect. A detected defect can also disappear or transform itself into another type of defect on the same rolling surface. For example, in many bearings the symptoms of a severe cavity on the outer or inner race were detected by periodic shock pulses. After several days of operation these symptoms disappeared to be replaced by the modulation of friction forces that characterize non-uniform wear of the same rolling surface. Detailed investigations that included disassembly of the bearings led to determining several causes for such changes.

Impurities transmitted by lubrication into the load zone of the bearing is one cause of this behavior. The symptoms of impurities within lubricating oil (external particles) before they are broken up and smoothed by the crushing action of the bearing coincide with the symptoms of a cavity on the corresponding raceway of the bearing. Pitting of the rolling surfaces is a second cause. After pitting, the process of smoothing the damage begins and the diagnostic symptoms of a cavity transform into the symptoms of wear and afterwards may disappear altogether. Repeated pitting in this zone may not reoccur until the end of bearing life. “Hardening” on the outer or inner raceway due to a large shock load on the machine rotor is a third cause of shifting defect symptoms. Again, the defect symptoms can disappear over time due to smoothing.

As stated earlier, about one third of the bearings studied had defects at the beginning of their operation due to faulty mounting of the bearing itself or a machine flaw such as shaft misalignment. At the end of bearing service life (MTBF) the number of faulty bearings does not increase significantly but the types of defects change. For example, when wear develops due to installation defects or excessive loads applied to the bearing the symptoms can significantly decrease and the defects themselves can disappear. The number of bearings with wear defects increases until at some time, that exceeds the mean service life (MTBF -- mean time before failure), reaches about 50% of all the bearings. At this time only about 20% have severe defects that limit the residual service life of the bearings. Finally, only about one-third of the bearings with severe defects are in serious enough condition to require immediate replacement.

The distribution of faulty bearings at the end of their service life, according the type of the most severe defects depends on construction, operating conditions and many other factors. Figure 9 shows two such distributions. The left chart is for machines with large horizontal rotors and a rotational frequency that does not exceed 10 Hz (600 rpm). The right chart is for medium to high speed rotating machines with vertical rotors and a load applied to the bearings that does not exceed 10% of the bearing's rated load. In both cases the machines were monitored and bearing condition assessed before general maintenance. The bearings in these machines have been in operation for about two MTBF. The segmentation in figure 9 represents the percentage of bearings found in conditions ranging from good to severe defects. It should be noted that some monitored bearings were replaced during the study without any records or accurate condition information. However, according to the users estimate, undocumented bearing replacement did not exceed 10-20% of the bearings that were monitored before replacement. Figure 10 shows the number of bearings replaced with different types of defects. Unfortunately, some monitored bearings were replaced during the study without any records or accurate condition information. However, according to the user's estimate, unassessed bearing replacement did not exceed 10-20% of the bearings that were monitored before replacement. As can be seen from the distribution graphs, slow speed horizontal machines with large rotors experienced outer (fixed) race defects most frequently. Inner (rotating) race defects were the primary defects experienced on vertical machines with light rotors.

Figure 9(above). Observed condition of bearings iin operation for about 2 MTBF.

Figure 10(above). Distribution of defects among bearings removed from two classes of machines.

An analysis of the bearings monitored periodically by the random vibration envelope spectrum method shows that there can be many models for defect development during machine operation. It is practically impossible to select one or two main models. Two models are often used in different industries. They are often used in different industries. One model represents a curve of the results of periodic condition assessment of bearings without installation defects. The intervals between the measurements are in MTBF. A total of seven defects is possible. There are two types of defects (wear and cavities) for each of the three rolling surfaces and one lubrication defect (increase in friction forces). Figure 11 most often describes horizontal machines with large rotors. Figure 12 most often describes machines with light vertical rotors. Two peculiarities of the behavior of these curves, not shown in the figure, should be noted. First is the possibility that some defects will repeatedly appear and disappear or transform from one type of defect (cavity) into another type (wear). The second is the relative stabilization of the type and severity of a defect when two or more developed defects are detected simultaneously in the bearing.

Figure 11(above). Time period between defect detection and bearing replacement for four types of defects - horizontal rotors.

Figure 12(above). Time period between defect detection and bearing replacement for three types of defects - vertical rotors.

Unfortunately there is no reliable information about the specific characteristics of defect development immediately before failure when there are several types of severe defects present simultaneously in the bearing. In Russia this lack of information about the final stage of bearing operation is due to the custom of immediate bearing replacement whenever two or more severe defects are detected in a bearing. The cases where a bearing with multiple defects was not replaced show that the increase in modulation of the random components and the increase in amplitude of certain harmonic component's may stop. During this last stage of bearing life immediately before failure, an increase in the magnitude of overall vibration and an increase in the level of components in certain frequency bands may be the only reliable indication of condition. Increasing vibration magnitude is typically observed right up to the point of failure.

Determining Bearing Lifetime

Statistical analysis of the results of periodic condition assessment requires some additional work to gain meaningful results. As an example, the maximum development rate of each defect type must be determined. To make a valid estimate, information was collected on bearings without installation defects that had been replaced. About 5,000 bearings were studied. Time to the appearance of an incipient wear defect and the type of defect were established. The time of bearing replacement and condition were determined. Bearings that had no severe defects at replacement were neglected. Bearings that were in a good condition and were still in operation at the time the study was finished were taken into account. About 60% of the bearings were in this latter category. Figures 13 and 14 show the data during the time period between defect detection and bearing replacement for different types of rolling surface and lubrication defects. From these data it is seen that rolling element defects develop at the highest rate but they are rare in comparison with other types of defects. Inner race defects rank next according to the mean rate of development. Lubrication defects are next in terms of the mean rate of development followed by outer race defects that typically develop at the slowest rate. Information about the rate of development of lubrication defects cannot be considered entirely reliable because fresh lubricant was often added to a bearing as soon as a lubrication defect was detected.

Figure 13(above). Number of defects noted by type.

Figure 14(above). Rate of defect development following detection.

This experiment was made on machines for which the MTBF of the bearings were not longer than three years. Operating time was estimated to vary in accuracy between 10% and 20%. As a result, the information about the rate of bearing defect development can differ slightly from the data illustrated in figure 14.

Long Term Lifetime Prediction Standard from Observed Data

The following will demonstrate that different methods for condition assessment and lifetime prediction are required during each stage of a bearing's service life. There is no single method that provides optimal results during all stages. Accurate condition assessment and lifetime prediction require separate solutions corresponding to the different stages of bearing lifetime:

Except for condition assessment and lifetime prediction during the final stage of bearing life, the preceding tasks are all accomplished from friction force excitation utilizing the high frequency, random vibration, envelope spectrum method of condition assessment. The envelope spectrum can also be utilized for the final stage condition assessment, but must be augmented by monitoring the increase in vibration level in wide frequency bands. Measuring the level of the bandpass filtered high frequency vibration prior to enveloping provides part of the information. However, this parameter primarily represents the condition of bearing lubrication. To eliminate possible ambiguity, measuring and trending the level of the bearing's middle frequency vibration is recommended.

An accurate prediction of long term bearing lifetime is necessary to establish a minimum period of normal, non-defective bearing operation. When no defects are present a probability of failure during the prediction period of about 0.01 should be used. This is equal to about 20% to 30% of the MTBF. A failure probability of about 0.03, equal to a value of about 10% of MTBF, should be used for estimating predicted lifetime in cases when a middle severity defect is detected. The operating period after a severe defect is detected can be determined with a failure probability of only about 0.1, close to 3% of MTBF. Finally, when a bearing has two or more severe defects it is close to failure and long term lifetime is impossible to predict.

The preceding determination of a predicted period for normal, non-failure bearing operation intervals relies on the absence of installation defects that significantly increase loads applied to the rolling surfaces. When installation defects are present, it is necessary to have statistical data defining the bearing MTBF for each type of defect. At the present time sufficient data is not available to make a reliable correction of the non-failure bearing operating life for each type of installation defect. However, selective data shows that a severe installation defect can decrease the predicted lifetime by about five times and a medium defect by about two times. Experimental data confirms the possibility of making a long term bearing lifetime prediction by the results of condition assessment, including an identification of the defect type, even if the defects are severe. Making a lifetime prediction requires determining the levels of the incipient, medium and severe defects. For each method of condition assessment, the rules describing how they are constructed and the levels corresponding to defect severity are determined by the designers of the diagnostic methods. In this case, a severe defect is defined as a defect when the probability of a bearing failure during operation is near the limit of acceptability. A medium severity defect is defined where the probability of a bearing failure during the period of lifetime prediction is low. Irreversible changes in condition that do not influence a bearing's serviceability or the time period of the lifetime prediction are defined as incipient defects.

It is possible to quote typical threshold levels for different defects as examples of condition assessment utilizing the magnitude of certain vibration components. Severe defects are usually 20-25 dB above the mean value of measurements obtained during initial condition assessment. Medium severity defects are generally defined when levels are about 10-12 dB above the mean. Threshold levels for incipient defects are not usually defined because the natural variation in periodic measurements of these components typically exceeds the value of the defect itself. In the random vibration envelope spectrum method, severity levels are based on the magnitude of modulation of the vibration signal. For high speed machines, about 20-25% modulation is considered indicative of severe defects. The percentage modulation considered indicative of a severe defect decreases in correspondence with a decrease in the bearing's rotating speed. Medium defects are usually defined at about one-half the percentage modulation established for severe defects. Modulation levels of incipient defects are defined by the sensitivity of the instruments used for condition assessment. If they are able to detect weak modulation of the vibration signal then the levels of incipient defects are set at one-third to one-fifth of the percentage modulation levels for severe defects.

Predicting Residual Lifetime Remaining During the Final Stage of Bearing Life

The process of condition assessment and lifetime prediction described above apply to the second stage of bearing service life, when wear defects appear and develop. However, there is a third stage of bearing life as defects increase to failure when it is still possible to estimate a bearing's remaining service life. The appearance of several developed wear defects defines the beginning of this stage in all cases, but the process to failure can be different. Defect development can progress in three principal ways during the third stage of bearing service life. In practice, one way often transforms into another.

The first type of progressive failure is characterized by a single, severe main defect which worsens throughout the service life of the bearing. The magnitude of modulation of the random vibration signal measured by the maximum amplitude in the random vibration envelope spectrum is a trendable parameter. In this case, the remaining service life can be estimated by projecting the severity trend of the maximum defect. Operating life remaining can be estimated from the magnitude of modulation, or its rate of change, or a combination of both these values.

The second type of progressive failure is characterized by the rapid appearance and development of new defects that quickly reach the severity of existing defects and determine the failure rate of the bearing. Under these conditions it is necessary to identify all the detected defects and construct trends for each individually or for the groups identified with each rolling surface in order to estimate remaining service life. This is necessary because defects of one group can transform into another. The magnitude of the diagnostic parameter associated with the most rapidly developing defect, or the rate of its change, or a combination of both these values is used to estimate the remaining service life of the bearing.

The third type of progressive failure is identified by a structural change in the vibration signal when a large number of defects appear in the bearing. Bearing condition is characterized by the appearance of high amplitude random time distributed shock pulse excitation. In the high frequency enveloped random vibration spectrum, the defect frequency components “widen” and essentially all the bearing diagnostic methods based on defect frequencies become ineffective. Under these conditions, mean vibration level and the rate of increase, the magnitude and rate of increase of vibration in wide, e.g., octave, frequency bands at middle and high frequencies or a combination of these parameters are the only practical methods for bearing condition assessment and estimating the remaining service life.

Optimizing the Condition Assessment System

Vibration measurement allows assessing bearing condition as well as the condition of other machine components. Optimizing the combination of diagnostic methods and means to include all likely problems is the most important challenge. This challenge is solved by considering the range and quality required for condition assessment, the price of the condition assessment system and the qualifications necessary for maintenance personnel.

The necessity to monitor rolling element bearing condition confronts the most machine users. Vibration measurements at fixed control points on the bearing housing are conventionally used for condition assessment. Using these control points for monitoring other components or the machine as a whole is often difficult because the vibration, measured at these points in the middle and high frequencies, is determined primarily by the bearing condition. Low frequency vibration components can be difficult to extract from a background of higher amplitude middle and high frequency components. In view of this factor, specially selected vibration control points on the bearing housing are recommended for efficient bearing condition assessment.

After selecting special control points for rolling element bearing condition assessment it is desirable to minimize the number of measurements without losing quality. The high frequency vibration envelope method is most efficient for rolling element bearing condition assessment. With this method 10 to 20 measurements will assure full condition monitoring and lifetime prediction for the entire service life of the bearing.

One operator with a portable system can monitor several thousand bearings with a high probability of detecting all defects long before failure. Thus, the necessity to include bearing condition assessment in an on-line machine condition monitoring system is a natural question. It is much more logical to include bearing condition assessment in on-line monitoring systems where failure can be sudden, costly, or the measurement point is inaccessible during normal operation.

The time required for one condition assessment measurement is a primary consideration that is determined by the frequency resolution of the analyzer. It must be about 10% of the shaft rotational period. Accurate condition assessment requires an average f not less than 8 to 110 spectra recorded over 80 to 100 rotor revolutions. For low speed machines (less than 60 RPM), the acquisition time can exceed two minutes. For most high speed machines, data acquisition will be less than one minute.

The next consideration is the time necessary for diagnostics. Existing automatic diagnostic systems require only a few seconds. In the small number of cases where operator involvement is required in the decision making process, the total time required is still only a few minutes.

Most have discovered that walking from point to point and attaching the transducer to the control point on the bearing housing consumes most of the time with portable condition assessment systems. In the simplest case, the transducer is fixed to a specially prepared surface by means of a magnet. To extend the band of frequency measurement, the point of attachment must be lubricated by a thin layer of grease prior to attaching the magnetically mounted transducer. Experience indicates that, under normal conditions, an attachment point can be cleaned, lubricated, the transducer attached magnetically, and a measurement recorded in less than two minutes.

In some cases, the optimum transducer location may be inaccessible. In this situation, the transducer should be attached permanently and the cable brought out to an easily accessible switch or junction box.

Conducting a full bearing condition assessment and lifetime prediction can be accomplished in less than 3 to 4 minutes per bearing. The total time required for condition assessment measurements during a bearing's lifetime is, to a large degree,determined by condition. Utilizing the enveloped random vibration spectrum assures the fewest number of measurements (about 10 to 20) on bearings that do not begin life with installation or their defects, are properly lubricated and operated within design limits. When condition assessment is accomplished by other methods, many more measurements may be required to construct an accurate comparative standard. A standard for anything other than the envelope method may also require revision following bearing or machine maintenance. Additionally, more frequent time intervals may be required between condition assessment measurements compared to the high frequency random envelope method to gain equal assurance. Therefore, the total number of vibration measurements and the time required with the envelope method may be as much as an order of magnitude less than would be required with any other method.

Operator training is an important factor that determines the efficiency and lifetime cost of a condition assessment system. When the assessment is made automatically by the random, high frequency envelope spectrum method, the operator typically does not need special knowledge and can be trained to operate the system successfully within several hours. Experience with the random envelope method has demonstrated that automatic diagnostics will allow a single person to monitor at least 5000 bearings. Without the automatic system, the number of bearings that can be monitored successfully by a single person decreases significantly.

Characteristics Required for Portable Systems Used for Bearing Condition Assessment

From the standpoint of cost, an instrument with bandpass filters, an envelope detector and spectrum analysis capability is an optimum portable condition assessment system for rolling element bearings. This instrument will perform the single measurement condition assessment described in the first part of this article and will gather all the data necessary for lifetime prediction. During periods where a series of measurements are required to establish the standards for lifetime prediction, following installation and repairs for example, temporarily connected on-line systems have advantages.

A condition assessment system for rolling element bearings can be an independent, application specific system, part of an overall machine condition monitoring system and operate either off or on-line depending on the criticality of the machines. When bearings are inaccessible, many users utilize permanently installed transducers connected to an accessible switch or junction box.

Due to the complex nature of high frequency vibration it is advantageous to perform the condition assessment and lifetime prediction automatically. An automatic system simplifies the work of the operator and does not require special training. A software program, DREAM, has been developed and proven in Russia. DREAM performs bearing condition assessment and lifetime prediction automatically and assures that the analysis is not affected or influenced by shock pulses transmitted from other components. The software will also enable an experienced user to accurately assess condition on complex machines. On machines of this type, such as gearing with rolling element bearings, shock pulses are generated by flaws within the bearings and on the gear teeth. Similar results can be achieved from mechanical transmissions that excite shock loads applied to the bearings.

With an accurate, efficient means to assess the condition of rolling element bearings it becomes much easier to identify and diagnose problems with other components.

A system for performing the high frequency random envelope spectrum method must have the following attributes:

Benefits of Bearing Condition Assessment

Implementing a condition assessment program for rolling element bearings provides the user with considerable economic benefits including:

Many of the largest enterprises in Russia have gained experience using portable diagnostic systems that automatically perform bearing condition assessment using the high frequency random vibration envelope spectrum method. These systems have typically provided a return of greater than 50%. In Russia, each operator is responsible for monitoring between 500 and 2000 bearings. This number could be increased. However, to do so it would be necessary to increase the number of permanently installed vibration transducers. Currently, permanently installed transducers are installed on less than 10% of the total monitored bearings. Permanently installed transducers are primarily installed on inaccessible bearings including some that require shutdown and even partial disassembly for access.

For some enterprises, the number of machine shut-downs in the time interval between scheduled maintenance has decreased by several times. Several enterprises have eliminated non scheduled shut-downs for bearing replacement altogether because all necessary bearing replacements are made during regularly scheduled maintenance.

Visual inspection of bearings and bearing replacement has decreased by one third. The number of bearings purchased has also decreased by one third. Expenses to assure step-by-step quality control of bearing fit and alignment on the shaft and in the bearing housing and the time and cost of bearing installation increase. However, these quality assurance expenses are less then the price of buying new bearings.

Bearings stocks are reduced as well. Requirements for replacement bearings are determined according to actual condition at least three months before they have to be installed. Practice has demonstrated that the cost of bearing condition assessment is quickly repaid by a substantial reduction in maintenance and lost production.


An assessment of vibration based rolling element bearing condition assessment and lifetime prediction during the different stages of operating life leads to the following conclusions:

  1. The operational life of a rolling element bearing can be divided in four main stages or periods according to condition: running-in, operation without defects, development of wear defects and bearing degradation.
  2. During each operational stage the optimal methods of bearing condition assessment and lifetime prediction utilizing vibration characteristics can differ significantly.
  3. A precisely located control point for vibration measurements on the bearing housing is required for accurate bearing condition assessment.
  4. During the first three stages of bearing lifetime (up to bearing degradation) it is possible to make an accurate assessment of bearing condition and lifetime. To accomplish this successfully it is necessary to detect and identify all the bearing defects that influence its operational life. At the final stage of bearing degradation, the remaining operational life is predicted mainly by the magnitude and rate of change of measured parameters.
  5. Bearing condition assessment and lifetime prediction does not depend on the methods and results of condition assessment used for other machine components. Bearing condition assessment can be accomplished with a single vibration measurement. Mastering the use of portable rolling element bearing condition assessment system should be the first stage of implementing a vibration based condition assessment program.
  6. high frequency envelope of the random vibration excited by friction forces and shock pulses in the bearing is the most effective and accurate and least expensive method of rolling bearing condition assessment. With this method, condition assessment can be accomplished confidently with only about 10 to 20 vibration measurements during the bearing's entire operational life.
  7. Algorithms and software for automatic condition assessment and lifetime prediction by the high frequency random vibration envelope method provide an accurate means to monitor a large number of rolling element bearings without a large amount of specialized operator training.

Return to Inteltek home page Return to articles/applications notes page