Peculiarities Of Slow Rotating Rolling Element Bearings Condition Diagnostics
Alexei Barkov, Natalia Barkova, Anton Azovtsev,
One of three primary methods is typically applied to solve practical problems of rolling element bearings condition diagnostics. The first uses the algorithms for defect detection using increase of temperature of the bearing housing, the second uses the detection of wear products in the lubrication oil while the third uses changes in vibration (noise) parameters. The most complete and detailed condition diagnostics of the bearings, including the detection and identification of possible defects even at the incipient stage of their development, uses the analysis of vibrations of the bearing housing, mainly the high frequency vibration components. The main problems in the application of this method appear when the high frequency vibration components are too weak, i.e., in slowly rotating machines and when the bearing housing is not accessible for vibration measurements. This paper presents the analysis of the condition diagnostics problems that occur in slowly rotating machines and solutions of these problems.
The main problems
Defects in the rolling element bearings can appear at three stages of the bearing life - during manufacturing, installation in the bearing shield, and during operation. Defects may also appear after transportation of the machine and its installation in the work place.
One type of typical manufacturing defect is the smooth deviations of rolling surfaces shapes from design specifications. These defects are most easily detected during the final bearing checkout on special stands using the low frequency vibration analysis methods. The vibration should be measured on the test stand supports. Of course, the rotation speed of the bearing in this situation should be relatively high, at least 3 to 5 Hz. This will insure easy and reliable low frequency vibration measurements. Another type of manufacturing defect may be hidden cracks and so fourth that lead to the appearance of shock pulses. For this type of defect detection, we recommend high frequency noise measurements in the "near" zone of noise emission. Using these methods in the bearing manufacturing quality checks, it is possible to avoid measurement problems caused by the slow rotation of the bearings during condition diagnostics.
The typical defects of bearing installation lead to significant increases of the local loads on the rolling surfaces and decreases of the lubrication layer thickness at the points where these loads are applied. These defects most often can be detected by the increase of middle frequency vibration components measured on the bearing housing although in slowly rotating machines, the vibration components excited by bearing installation defects are found in the lower frequency domain. In this case, the problem occurs of how to distinguish between vibration components excited in the bearing shield and the components that came from other machine units or even other machines. Also, one must face the problems connected with the low level of the bearing vibration. In the first approximation, the bearing vibration level is proportional to the square of the rotation speed. Another problem is the frequency resolution required to resolve the bearing frequency components that may differ by the cage rotation frequency. These problems can be solved by using measurement instrumentation with very high dynamic range and high frequency resolution including zoomed spectra techniques.
The defects that occur during rolling element bearing operation, first of all, influence the properties of friction forces in the bearing. That's why in the incipient stage of such defects development, they can be detected only by the analysis of high frequency vibration excited by friction forces. The shock pulses that appear when rolling elements contact the defective surfaces can be considered as one of friction force components as well. They can be also detected by the analysis of high frequency vibration. These types of defects can significantly influence the low frequency vibration of the machine only when they reach dangerous stage. Before that, they can not be reliably detected using the low frequency vibration analysis.
A number of problems can occur during bearing condition diagnostics in slowly rotating machines. These problems require application of more complex methods for diagnostics and advanced measurement and signal analysis instrumentation. The reason for this is the low level of high frequency vibration in bearings of slowly rotating machines and in addition the high level of low frequency vibration components that propagate to the measurement points from other machine units and even other machines and equipment mounted on the same foundation. As the high frequency vibration levels decrease with the decrease of the machine rotation speed much faster than low and medium frequency vibration levels, diagnostic methods using the high frequency vibration excited by friction forces require, not only the high sensitivity from the measurement instrumentation, but also the extension of such parameters as dynamic range and linearity as well. Only in this case it is possible to extract the high frequency vibration components from the background of the instrumentation natural noise and the distortions of the low and medium frequency signal processing. We should note that low and medium frequency vibration components in slowly rotating machines typically exceed the level of high frequency vibration by a factor of 1000 or more. For the condition diagnostics of bearings using medium frequency vibration which is the case when the high frequency vibration can not be measured or after the bearing installation, instruments with very high frequency resolution are necessary.
Methods of Diagnostics
There are two main symptoms for the detection of incipient bearing defects, the appearance of shock pulses and the appearance of amplitude modulation of friction forces within the bearing. In the bearings of slowly rotating machines, the lubrication layer is thinner than in faster rotating machines and, for this reason, the lubrication layer may be more easily broken which causes shock pulses to occur more often. The amplitude of shock pulses in slowly rotating machines is much lower, but their number significantly exceeds the number of shock pulses in the normal machines which results in the overlapping and integration of shock pulses in time. In practice, we can consider that this mechanism produces "dry" rolling friction forces. Even if we manage to extract the high frequency vibration of a bearing with natural wear, it will be extremely difficult to find the components excited by shock pulses. It means that the shock pulse methods for the diagnostics of rolling element bearings in slowly rotating machines can be applied only in rare special cases.
The mechanism of friction force amplitude modulation in defective bearings is the same for both slow and fast rotating machines. Furthermore, this is true despite the fact that friction in high speed bearings has a hydrodynamic origin while in slow speed bearings, the friction has mechanical or at least combined mechanical and hydrodynamic origin. This is why the method of spectrum analysis of the random vibration power oscillations, well known as the enveloping method, can be applied for the condition diagnostics of both slow and fast rotating machines' bearings.
Figure 1 illustrates the basics of the enveloping method. Here, you can see high frequency vibration components, their envelopes, and envelope spectra of good and defective bearings. In a good bearing, the signal power does not have any periodic low frequency oscillations and, in a defective bearing, these oscillations can be significant and can be easily detected in the envelope spectrum by the presence of harmonic components with bearing frequencies and their combinations. The defect type is identified by the frequencies of the harmonic components found and the defect severity is identified by the modulation index, the relationship between the amplitudes of the harmonic and random components in the envelope spectrum [1,2].
The enveloping method allows the distinction between the two groups of each type of bearing surface defect developed during operation. The two basic types are wear which is characterized by the smooth modulation of friction forces and cracks or spalls which are characterized by sharp changes in the friction forces. This method also allows the detection and identification of the types of dynamic loads applied to the bearing by the rotor. This can also be used to detect jointed coupling defects including the defects of geared couplings and mechanical transmissions including both gear and belt ones [3,4]. The list of bearing defects detected by the analysis of dynamic loads includes such bearing installation defects as race misalignment, high (nonuniform) radial or axial tension, and others.
a. High frequency vibration components. No modulation or shock pulses present
b. High frequency vibration components of a signal with amplitude modulation
c. Envelope of signal from "a" plot
d. Envelope of signal from "b" plot
e. Envelope spectrum of signal from "a" plot
f. Envelope spectrum of signal from "b" plot
|Figure 1.High frequency vibration components of the signals with amplitude modulation and without it. "a, b" - time domain plots, "c, d" - envelope time domain plots, "e, f"- envelope spectra|
The full list of defects detected by the envelope spectra analysis can be divided into three groups. The first group is the bearing installation defects:
The second group includes defects developed during bearing operation:
The third group is formed by defects of the rotor, jointed couplings, and transmissions that produce dynamic loads on the bearings including:
Fig. 2. Vibration spectra of one and the same machine rotating at different speeds - 25, 10 and 3 Hz.
The problems of rolling element bearings condition diagnostics are mostly connected with the difficulties in high frequency vibration measurements. In high speed bearings, the primary difficulty is the loss of high frequency vibration power with the high frequency vibration propagation. This case can be solved by to mounting the vibration transducer on the elements of bearing housing that are in direct contact with the stationary bearing race. In slowly rotating bearings, the relationship between the low or medium frequency vibration level and that of the high frequency components is much worse than in the high speed machines and this is the reason for additional problems which include the separation of the low and high frequency vibration components and the extraction of the high frequency signal components from the internal noise of the measurement instrumentation including the transducer.
Figure 2 illustrates that the high frequency vibration level decreases much faster with the decrease of the rotation speed than the levels of low frequency vibration. The figure presents the vibration spectra of the bearing housing of the machine rotating at 25, 10 and 3 Hz. That's why, in the case of slowly rotating machines, even the use of spectrum analyzers with high dynamic range and high linearity parameters does not always allow the measurement of high frequency vibration components in a satisfactory manner. In this case, we must analyze the power oscillations of the signal at lower frequencies as illustrated in figure 2. The main problem in this case is the selection of the frequency band for the analysis or the use of appropriate signal processing techniques to avoid the presence of strong harmonic components and narrow resonances in the frequency band being analyzed. In the opposite case, the presence of two or more harmonic components in the demodulated signal will produce their combination harmonics in the envelope spectrum and in most cases it will not be possible to separate these harmonics from the ones that reveal bearing defects. If the analyzed frequency band contains even a single, very strong harmonic component, then the amplitudes of signal power oscillations decrease and the information about the defect severity becomes incorrect. To eliminate these type of errors in the condition diagnostics of bearings, we strongly recommend making autospectrum measurements and analyzing these with the aim of selecting the proper frequency band for enveloping .
Currently, the authors have obtained substantial experience in the condition diagnostics of slowly rotating bearings. The slowest rotating bearings were from a number of machines in metallurgical industries which had considered rotation speeds about 5 to 7 revolutions per minute. Usually, the main problem in this situation was the limited time when the bearing was operating under stable load and speed.
To obtain accurate condition diagnostics of a bearing using enveloping methods, it is necessary to make vibration measurements over a period of 50 to 100 revolutions of the bearing. In other industries such as power and paper where the machines operate with stable rotation speeds for a long period of time, there is no need for the introduction of special modes of bearing operation or measurements. All the peculiarities in bearing condition diagnostics of slowly rotating machines in these industries are connected with the proper choice of the frequency band for envelope detection and the introduction of additional symptoms for the detection of rolling elements defects. Figure 3 presents envelope spectra of bearings with rolling element defects in high and low speed machines. From here, the necessity for changes in the symptoms list becomes evident. In the second spectrum, you can see the "widening" of the spectrum components related to the cracks and spalls of the rolling elements. One more problem for all industries is the condition diagnostics of the bearings in output stages of reduction gearboxes. The problem is that these bearings are affected by the high speed stages vibration of the gearbox. In this case you, should be very careful in choosing the vibration transducer location and frequency band for enveloping.
Fig. 3. Envelope spectra of the bearings of electric motor and drying cylinder of the paper machine. Both bearings have similar defects of the rollers. Here RPM is the rotation speed of the shaft, BSF - ball spin frequency, FTF - cage rotation frequency, Fz - gearmesh frequency of the drying cylinder drive.
Fig. 4. Envelope spectra of the bearing vibration at the third stage of the gearbox. The first spectrum is selected by a 3.15 kHz 1/3 octave bandpass filter, the second - by a 10 kHz 1/3 octave bandpass filter. Here RPM1 - rotation speed of the first stage, RPM3 - rotation speed of the third stage, BPFO - ball pass frequency outer race of the bearing at the third stage. The vibration was measured at this bearing housing in both cases.
Figure 4 presents envelope spectra of the rolling bearing of the output stage of a three stage reduction gearbox. The rotation speed of the bearing is 1.4 Hz. The spectra are measured in the frequency bands of 3.15 kHz and 10 kHz. In the first case, you can easily see the symptoms of the first stage gears defects which reduce the modulation index for the bearing outer race cavity symptoms. The second spectrum presents only the third stage bearing defect symptoms.
The limitations to this technique, caused by the limited time when the bearing operates under stable load and speed, are most often found in the metallurgical industry and especially in rolling production when the load on the press is stable only for a few revolutions and in lifters when the time available to lift something is also limited to a few revolutions of the bearing. Three main approaches to the necessary vibration measurements, including the high frequency vibration envelope, can be considered under these limitations.
The first approach can be considered as the main one. It is the use of step by step data acquisition. If we can find a time interval when the bearing makes at least 3-4 revolutions with stable speed, we can make successive envelope spectrum measurements and then average them with similar measurements in the next time intervals.
The second approach to be considered is the introduction of special diagnostic modes of machine operation. One case is operation under no load but with stable speed. A first example is condition diagnostics of aircraft jet engines when the engine is rotated either by the starter or by an external drive. Another example is the condition diagnostics of box bearings in the railway locomotives when the locomotive is lifted and the bearings are rotated using the main motors of the locomotive. The problems in this type of diagnostics are connected with the fact that, during such measurements the defects of the stationary bearing race may not be loaded, especially in case of to high bearing clearances and a defect may be missed. An opposite situation is possible as well. For example, if the newly loaded points coincide with the point where in the normal operation of the bearing, the lubrication waste was gathering. Then, in the first minutes of measurements, we can see false information of the presence of defects. Figure 5 presents such an illustration. Here, you can see an envelope spectrum measured on the box bearing of the lifted locomotive. In the lower part of the bearing where the load is applied after the locomotive was lifted, some lubrication leftovers were accumulated during normal operation plus some incipient rust occurred. During the first minutes of rotation, the envelope spectra presented the symptoms for cracks but, after 10 minutes, only symptoms for race wear were found, and after 30 minutes, all symptoms for defects disappeared.
The examples allow the conclusion that, when you choose a special operation mode for machine diagnostics, it is extremely important to preserve the load point and direction applied to the bearing in normal operation mode.
The third approach to the condition diagnostics of bearings operating under continuously changing rotation speed is the synchronous analysis of the spectra using the shaft rotation sensors which is sometimes called order analysis. The practice shows that using this method, it is possible to diagnose bearings even in oscillating mechanisms, in particular, the bearings in converters in metallurgical applications. But, there are also significant problems if the rotation speed during measurements changes in more than 15 to 20%.
Fig. 5. Envelope spectra of the locomotive box bearing vibration.
The most important characteristics of the condition diagnostics quality are the probabilities of missing severe defects, false alarms, and errors in the exact defect type identification. The first two probabilities define the condition diagnostics efficiency and the third one defines the efficiency of long-term condition prediction. The condition prediction interval is defined by both the developing defect types and their severities.
Experimental results of bearing defect identification after diagnostics shows that quantitative estimations of the above probabilities are very close for both slow and fast rotating machines. Mainly, they are defined by the selected sequence of the diagnostic measurements.
The best results, an error level of only 3% for the sum of defect missing and false alarm probabilities, are achieved if the high frequency vibration envelope spectra and medium frequency autospectra are measured periodically on the bearing housing. Time intervals between measurements of about 10% of bearing MTBF are sufficient to make a long term safe operation time prediction and prediction of residual service life after the well developed defects are detected in the bearing. The latter is achievable by using trends of low and especially medium frequency vibration components.
High efficiency is also obtained using a single measurement of the high frequency vibration envelope spectrum and the vibration autospectrum and further comparison of the results with similar measurements on groups of identical machines. In this approach, we can say that decreases occur mainly in the reliability of residual service life prediction in the presence of severe defects, but the reliability of condition diagnostics and safe operation time prediction when the defects are at lower stages of development is the same as in the first case.
The third method of the condition diagnostics of slowly rotating rolling bearings which uses a single measurement of high frequency vibration envelope spectrum has lower reliability, but it still is efficient in more than 90% of practical problems. Two possible errors can be done in condition diagnostics using a single measurement. The first potential error is missing lubrication defects that lead to an increase of friction forces and high frequency vibration level and the second problem is that it may be impossible to separate the groups of severe defects in the bearing in near to failure conditions when the rolling elements do not rotate periodically and friction forces are not modulated periodically, but only at random. However, in both cases the defective bearing can be detected from higher vibration (noise) levels and increased temperature.
Reliable condition diagnostics of slowly rotating rolling bearings requires making vibration envelope spectra at frequencies at least 500 to 1000 times higher than the rotation speed. Taking into account that the vibration levels at these frequencies is typically less than 1% of the overall vibration level, it is evident that the analysis of these signals requires that the selective band pass filters in the envelope detector should be of at least 6th order with a dynamic range of at least 70dB with very high linearity. So, in order not to complicate and increase the price for measurement instrumentation, it is worth using digital filters and envelope detector. In modern instrumentation developed with the participation of the authors, specialized digital signal processors are used for this purpose.
Currently, in the vibration measurement and analysis instrumentation used for rolling element bearings condition diagnostics, two types of signal analysis are available prospective. These are normal spectrum analysis and high frequency envelope analysis of signals selected by a bandpass filter. These types of vibration measurements are sufficient for condition monitoring of a machine and for condition diagnostics of bearings as well as other units of a rotating machine .
Measurement instrumentation for rolling element bearings condition diagnostics are produced in three main types which include types based on PC hardware and appropriate software, special measurement blocks with several vibration sensors, and portable signal analyzers and data collectors. These instruments are discussed in more details in other papers [6,7].
The peculiarities of the signal analysis techniques used for rolling element bearings condition diagnostics discussed in this paper in combination with modern methods for the automatic state recognition allow making systems for automatic condition diagnostics of rolling element bearings. These systems are widely used in Russia and provide, not only reliable detection and identification diagnostics of rolling element bearings defects, but also long term condition prediction as well.