Paper C Miettinen, J., Pataniitty, P. Acoustic Emission in Monitoring Extremely Slowly Rotating Rolling Bearing. In: Proceedings of COMADEM 99. Oxford, Coxmoor Publishing Company. 1999. ISBN 1-901892-13-1. pp. 289-297.
1 Acoustic Emission in Monitoring Extremely Slowly Rotating Rolling Bearing Juha Miettinen * and Pentti Pataniitty ** * Tampere University of Technology, Machine Design P.O. Box 589, SF-33101 Tampere, Finland email: miettinen@ruuvi.me.tut.fi ** Acutest Oy, Hermiankatu 8 D SF-33720 Tampere, Finland email: acutest@co.inet.fi Abstract: There are not international standards or universally accepted limit values available, that classify rotating machines as slow or high speed machines. The old international standard ISO 2372, which has been replaced with the new ISO 10816 standard, gave the vibration velocity severity ranges for different classes of machines. The old standard covered machines with rotational speeds from 600 rpm to 12000 rpm. The new standard does not contain any rotational speed limits. Sometimes the limit for low-speed rotating machines is set to 20 rpm or 30 rpm. In industry, it is easy to find machinery where the rotational speed in continuous running is lower than 2 rpm. In the condition monitoring of rotating machines, it is common practice to measure the vibration velocity or acceleration. At very low frequency, the vibration velocity amplitude becomes weak and therefore displacement measurement can sometimes be a suitable vibration measurement parameter. When the rolling element in the rolling bearing passes the early-stage fault in the case of an extremely low rotational speed, the energy that the collision generates is very low. In that case, the defect is difficult to detect in the frequency domain but can possibly be seen in the time domain. The frequency bandwidth of acoustic emission (AE) measurement method is typically in the range 100 khz to 1 MHz. In that range, vibrations occur in a material by fracture of crystallites, crack nucleation and growth, several mechanisms involving dislocations, phase transformations in materials, boiling and electrical discharges. Each of these mechanisms is characterised by a rapid collective motion of a group of atoms. The present paper describes the use of the acoustic emission method in the monitoring of faults in an extremely slowly rotating rolling bearing. The introduction describes the principle of the measurement method of acoustic emission and the analysis methods used for the acoustic emission signal. The paper contains the results of AE measurements where the rotational speed of the shaft was from 0.5 rpm to 5 rpm. The measurements were carried out using a laboratory test rig with grease lubricated spherical roller bearings of an inner diameter of 130 mm and a load of 70 kn. Prior to testing the test bearing had been naturally damaged on its outer race during normal use in industry. The results of the acoustic emission measurement have been compared with the results of low-frequency vibration measurements, which have been carried out in the same test arrangement. The paper gives an example where acoustic emission measurements have been used in industry, in the monitoring of slowly rotating machinery. Keywords: acoustic emission, vibration measurement, slowly rotating bearings 1. Introduction Acoustic emission can be described as a shock wave inside a material, which is under stress. The shock wave causes the surface of the material to move, and this movement is measured with a very sensitive sensor. The shock, or transient elastic wave, is generated by a rapid release of energy from a local source within the material. The sources of acoustic emission (AE) comprise different mechanisms of deformation and fracture including the fracture of crystallites, crack nucleation and growth, several mechanisms
2 involving dislocations, phase transformations in materials, boiling and electrical discharges. Each of these mechanisms is characterised by a rapid collective motion of a group of atoms (Beattie, 1983). The intensity of acoustic emission vibration, often called AE activity, depends on the type of material and on the properties of the material. In Table 1 (Miller, 1987), some factors that affect the relative amplitude of the acoustic emission response are presented. Table 1. Factors that affect the relative amplitude of the acoustic emission response (Miller, 1987). Factors that tend to increase the acoustic emission response amplitude High strength High strain rate Low temperature Anisotropy Nonhomogeneity Thick sections Brittle failure (cleavage) Material containing discontinuities Martensite phase transformations Crack propagation Cast materials Large grain size Mechanically induced twinning Factors that tend to decrease the acoustic emission response amplitude Low strength Low strain rate High temperature Isotropy Homogeneity Thin sections Ductile failure (shear) Material without discontinuities Diffusion-controlled phase transformations Plastic deformations Wrought materials Small grain size Thermally induced twinning A particular feature, which affects the activity of the acoustic emission, is called the Kaiser effect. The Kaiser effect means that when a defined stress has been applied on the material and it has caused acoustic emission, additional emission will not be induced in to the material until defined level of stress has been exceeded, even if the load is completely removed and then reapplied (Miller, 1987). Because of the Kaiser effect, each AE wave can occur only once. The Kaisen phenomenon has a special effect on the crack nucleation and growth. It can also affect the AE activity caused by a fault in a bearing when a rolling element is passing the fault and causes stress in the material. In order to evaluate the significance of an AE source and to interpret the AE signal, different parameters can be extracted from the signal. The signal waveform depends on the characteristics of the source, on the path of the signal from the source to the sensor, on the characteristics of the sensor and on the measurement system. The parameters, which will be extracted from the signal, are depending on the type of the signal. From a burst type of signal, typical extracted parameters are the duration time of the AE event, the emission counts, the emission event energy, the emission signal amplitude and the peak amplitude, the emission signal rise time or the signal decay time, see Fig. 1. Fig. 1. An AE burst signal and the characteristic parameters of the signal. If the examined AE signal is longer and contains a lot of emission bursts like the signal shown in Figure 2, one way to characterise the signal is to calculate statistical values from the time signal. These values express if the signal is peaked or not. Typical statistical values are the standard deviation of the signal, the kurtosis value of the signal, the variance value of the signal, the skewness value of the signal, the signal peak-to-peak value and the acoustic emission signal root mean square (RMS) value that describes the signal energy
3 content (Li, 1995). One way to characterise the signal is to count the pulses per time unit; for example pulses/second. This way to monitor the signal is very suitable for continuous measurement because the data, which is stored, contains only a fractional part of the amount of the data that is stored if the AE time signal is measured. The pulse count data is easier to handle in the industrial environment. In some studies an AE time signal area summation technique has been used (Tan, 1990). Fig. 2. A long AE time signal, which contains a lot of emission bursts. 2. Acoustic emission in the monitoring of faults of rolling bearings When the rolling element passes the fault in a bearing, it excites vibration, which can be measured by a vibration measurement sensor. The common practice is to measure the vibration displacement, velocity or acceleration. The velocity amplitude is almost independent from the vibration frequency in the range 10 Hz - 2 khz. The vibration velocity describes very well the general condition of the rotating machine. For that reason, vibration velocity measurement is the most common measurement parameter in condition monitoring of rotating machinery. At higher frequencies, the vibration displacement amplitude becomes very low but the vibration acceleration rises to a high level. For machines rotating at high speed with a rolling bearing failure at a very early stage, the measurement of the vibration acceleration is usually a more reliable indicator (Berry, 1991). The type of the spectrum, which the defective rolling bearing generates, depends on the severity of the fault. We can find out basically four types of spectrums. These spectrums can contain random ultrasonic frequencies, natural frequencies of bearing components, bearing rotational defect frequencies and sum and difference frequencies which are born when the different frequencies modulate with each other (Berry, 1991). Random ultrasonic frequencies appear typically at an early stage of the fault, and can dramatically rise just before the seizure of the bearing. The natural frequencies of the mounted bearing are utilised when vibration acceleration is measured using enveloping techniques. The main difference between the acoustic emission and the low frequency vibration is that for low frequency vibration we already need a defect in the bearing which excites the vibration when a rolling element is passing it but acoustic emission vibration is exited just when the crack is formed. Therefore, in principle, the acoustic emission should indicate damage in the bearing at a very early stage. In practice, the situation of course is not so clear. When the rolling bearing is running, the lubrication situation is not always fully flooded but very often the bearing is running under some degree of starved condition. Starved lubricated situation means metallic contacts, micro fractures, and micro plastic deformations, which all can generate acoustic emission. 2.1 Technique to measure acoustic emission The substantial difference in the AE measurement technique, compared with the low frequency measurement techniques, is in mounting the measurement sensor. Acoustic emission monitoring is nondirectional. Most AE sources appear to act as point sources. The point sources radiate energy in spherical wavefronts, and therefore the sensor can be located anywhere in the vicinity of the AE source and it can detect the acoustic emission signal. This is in contrast to other measurement methods of mechanical vibration, in which the direction of the sensor has a strong influence. The vibration measurement sensors are very insensitive in other directions than the measurement direction so as result we get the vibration in the direction of the measurement axle of the sensor.
4 Although the AE measurement is nondirectional, it must be taken into consideration that every boundary surface affects damping on the high frequency vibration signal. Therefore, the sensor should be located as close as possible to the expected emission source, normally on the load side of the bearing. In addition, it is very important to use a contact grease between the sensor and the fitting surface. Because the levels of the voltage in the acoustic emission measurement are very low, great care must be taken to minimise the affect of disturbances from the environment on the measurement. Strong electrical disturbances can cause, for example, magnetic fields, eddy current fields, inverters of the electrical motors and fluorescent lamps. The so called background noise, which means the acoustic emission generated for example from pressure vessels, welding, hydraulic and mechanical noise, fretting and deformations by heat expansion, can disturb the AE measurements, especially in field environments. The disturbances can affect the measurement results, they can not always be easily explained. The disturbances are especially harmful, when the rotational speed of the bearing is low and when the emission level from the bearing is low (McFadden, 1984). The background noise can affect the AE measurement in that way, that even if the signal is very clear when the rotational speed is low the signal can become ambiguous when the rotational speed and the background noise is higher (Smith, 1982). The frequency bandwidth in acoustic emission measurement is typically in the range 100 khz to 1 GHz. The sensors are generally of piezoelectric type. The difference between the normal accelerometer and the AE sensor is that AE sensor does not have any mass attached on the piezoelectric crystal. The frequency response of the AE sensor is strongly non-linear and therefore the measurement of the spectrum in the case of acoustic emission is not very suitable. In Figure 3 is a graph of a typical frequency response of an AE sensor presented (B&K, 1984). The normal way to do the measurement is to use a narrow band-pass filter which centre frequency is the same as the resonance frequency of the sensor. Fig. 3. A typical frequency response of an AE transducer (B&K type 8313). 3. An example of the acoustic emission monitoring of slowly rotating rolling bearings in the paper industry The example is from a Finnish paper plant. The AE monitoring system is installed for monitoring the support bearings of the lime sludge reburning kiln, Fig. 4 a and Fig. 4 b. The kiln is supported with six bearing pairs, total number of the bearings being 12. The rotational speed of the kiln is 8 rpm. During 10 years running there has happened over 16 bearing faults, which have caused unsystematical shutdowns and losses of production. The principle of the AE measurement in this case is continuos pulse count method. The AE sensors were of piezoelectric type and their sensitivity was highest at the frequency of about 150 khz. The lower frequencies until 100 khz are filtered out. After some time the measurement system was taken in to use the pulse count level from the bearing number 8 started to rise from its normal level. That situation is shown in Fig. 5 a. After running of 20 days the pulse count level started to rise again very strongly which is shown in Fig. 5 b. The kiln was stopped according to the normal maintenance plan and the bearing was removed. That maintenance operation did not affect any costs in the production. After the removal of the bearing the pulse count level dropped to a level about 100 pulses/300s.
5 a) b) Fig. 4. The lime sludge reburning kiln (a) and the supporting bearing (b). a) b) Fig. 5. The AE pulse count level of the bearing number 8 starts to rise slowly (a) and after 20 days operation the level started to rise very strongly (b). 4. The measurements in the laboratory The aim of the laboratory measurements was to test the acoustic emission measurement with extremely low rotational speed of a rolling bearing. For the sake of comparison, some low frequency vibration measurement methods were included in the measurements. These methods were the envelope spectrum method, the peak value method, the method of derivation of the acceleration signal and the time signal of vibration acceleration. 4.1 Measurement arrangement The measurements were carried out with a test rig, which is shown in Fig. 6. The type of the test bearings and the support bearings of the rig was two-row spherical roller bearing. The bearing application was the same as in a normal railway wagon. Every housing includes two bearings and one of the test bearings was damaged. The damage was followed from the normal use of the bearing and the damaged bearing was found in normal maintenance operation of the wagon bearings. The test bearing represents a typical faulted bearing in that usage. The load of the test bearings during the measurement was 70 kn per housing and it was static and pure radial. That load is the same as the maximum static load per one housing when the wagon is loaded full. The type of the grease in the test bearings was NLGI grade 1.5 lithium complex soap with synthetic base oil. The AE sensor was mounted on the bearing housing with a 150 mm long wave-guide, which was fitted with screw fastening. The AE sensor was of piezoelectric type and the signal was filtered with narrow band-pass filter which centre frequency was 150 khz or 240 khz. The methods for analyse of the AE signal were the pulse count method and the time signal of acoustic emission vibration. The pulses where counted in the unit of pulses/one second. From the time signal, the cycle time of the fault frequency of the bearing was defined when it was possible. The rotational speed in the measurements was from 0.5 rpm to 5 rpm. With low frequency measurement methods, the rotational speed was so low when the fault still could be detected.
6 Fig. 6. The test rig of the bearings in the laboratory. 5. Results In the results are shown the AE pulse count and the AE time signal measurements when bearing rotational speed has been 0.5 rpm, 0.85 rpm, 1.4 rpm and 5 rpm. In the end of this chapter, also some results of the measurements of low frequency vibration are shown. In generally the fault was identified in all AE measurements. In some of the cases, the background noise appeared so strongly that it was difficult to find out the cycle time of the fault. In Fig. 7 are shown the results when rotational speed was 0.5 rpm. The fault cycle time can be identified very clearly from the time signal and from the pulse count results. The background noise with that a slow rotational speed is low comparing it for example between the results when the rotational speed was 0.85 rpm, which is shown in Fig. 8. Because of the low background noise, the fault cycle time can be identified more clearly from the AE time signal measurement when the rotational speed was 0.5 rpm than 0.85 rpm. The time of the pulse count measurement is much longer than the time of the AE time signal measurement. From the long-time pulse count results it is possible to find out time intervals where the cycle time of the fault can be easily identified despite of the background noise, which can be seen from Fig. 8. AE pulse count AE time signal Fig. 7. The AE pulse count and the AE time signal results with a rotational speed of 0.5 rpm. AE pulse count AE time signal Fig. 8. The AE pulse count and the AE time signal results with a rotational speed of 0.85 rpm.
7 In Fig. 9 the AE pulse count and time signal results when rotational speed was 1.4 rpm are presented. The background noise in these measurements was so high that it was a little bit difficult to find out the cycle time of the fault frequency from the AE time signal. The background noise disturbed also the pulse count measurement and therefore the cycle time could not be identified as well as in that case when the rotational seed was lower, which is shown in Fig. 7. AE pulse count AE time signal Fig. 9. The AE pulse count and the AE time signal results with a rotational speed of 1.4 rpm. In Fig. 10 the AE pulse count and time signal results when rotational speed was 5 rpm are presented. In this measurement, the emission from the fault is so strong that the background noise does not disturb the measurement. The cycle time of the fault frequency can be identified very clearly from the time signal. The AE pulse count was measured in the unit pulses/one second. When the cycle time of the fault frequency approaches the time interval of the pulse count measurement, the cycle time of the fault can not be anymore identified from the pulse count results. This is shown in Fig. 10. In the pulse count result the cycle time of the rotation of the shaft can be seen very clearly but the cycle time of the fault frequency can not be identified. AE pulse count AE time signal Fig. 10. The AE pulse count and the AE time signal results with a rotational speed of 5 rpm. The character of the AE emission depends for example on the size and on the shape of the fault on the rolling surface. In Fig. 11 is shown one AE pulse cluster, which has been taken from the result of the time signal measurement shown in Fig. 10. From the fault in the bearing, a pulse cluster is generated, which in this case consists of eight short-duration AE bursts. This kind of pulse cluster has a different shape than for example the acceleration signal in the low frequency area generated from the same fault. Fig. 11. The AE pulse cluster from the fault in the bearing.
8 That is one difficulty in the pulse count technique or for example in the peak value technique in analyses of the AE signal. If the cycle time of the fault is short, it may be difficult to find out an adequate time interval for the measurement so that the different clusters will not be mixed together in the measurement. That makes it difficult to identify the passing frequency of the fault. One possibility in this case is to follow the AE pulse count overall level or the AE activity RMS value. 5.1 Results of the low frequency measurement The same damaged bearing was measured also with the low frequency measurement methods. The methods, which were included in the measurements, were the envelope spectrum method, the peak value method, and the method of derivation of the acceleration signal and the pure time signal of acceleration. Some view of the suitability of the low frequency measurement methods in this case is shown in Fig. 12. In these measurements, the most sensitive low frequency measurement method was the envelope based spectrum measurement method. In Fig. 12 methods have been arranged based on the property of detecting the fault from the spectrum and from the time signal. The grades have been estimated. Envelope method was scored as a highest, because it could detect the fault in a lowest rotational speed. The limit of the rotational speed when the fault was detected was with envelope method 10 rpm and with the other methods 20 rpm. Fig. 12. The suitability of low frequency measurement methods to detect the fault in the bearing. The methods are arranged based on the property of detecting the fault from the spectrum and from the time signal. The grades have been estimated. Envelope method was scored as a highest, because it could detect the fault in a lowest rotational speed. Conclusions The acoustic emission measurement has been tested to detect the fault of a rolling bearing, which is rotating with extremely slowly rotational speed. The rotational speed in the measurements has been from 0.5 rpm to 5 rpm. The study denoted that the acoustic emission measurement is a very sensitive method to detect the fault in a bearing which is rotating with an extremely slowly rotational speed. With the AE method, the fault in the bearing could be identified with the slowest rotational speed, which was used in the measurements. With the lowest speed the fault was clearly identified from the AE time signal and from the results of the AE pulse count method. The voltage levels in the AE measurements are very low therefore, the influence of the external disturbances on the measurement must be taken in the consideration. The external disturbances can be caused by other emission sources than from the rotation of the bearing. In addition, the high rotational speed can cause background noise, which can make it difficult to identify the cycle time of the fault frequency. The pulse count method is convenient principle for monitoring the rolling bearing when the rotational speed is extremely slow. The measurement time in pulse count is long, very often continuous. From those long time measurement results, it is possible to find out time intervals of that kind from where the cycle time of the fault frequency can be clearly identified. The pulse count method has also the advantage that the size of
9 the measurement file stays reasonable. Instead, in the measurement of the AE time signal the data file can grow up so large that it is troublesome to process. When the rotational speed was higher, the cycle time of the bearing fault frequency could be identified best from AE time signal. With higher rotational speed, the collision of the rolling element on the fault creates so high emission energy that the background noise does not disturb the identification of the cycle time of the fault. The measurement time interval in the pulse count method is limiting its use in identifying the fault cycle time when the rotational speed is high. The character of the AE emission is depending for example on the size and on the shape of the fault on the rolling surface. When the rolling element passes the fault it builds up an emission burst cluster. In this case, it is difficult to choice the right time interval for pulse count method and it is possible that the different emission clusters are mixed together. The measurements were carried out also with low frequency measurement methods. The limit of the rotational speed when the fault still could be reliable identified with those methods was between 10 rpm and 20 rpm. Acknowledgements This study is a part of the Finnish project Condition Monitoring of Grease Lubricated Rolling Bearings" which is included in the international COST 516 GRIT research programme. The authors are grateful for the financial and technical support from the following companies and institutions: The Technology Development Centre of Finland (Tekes), SKF Engineering & Research Centre B.V. in The Netherlands, the Finnish companies Mobil Oil oy ab, Rautaruukki Steel, VR Ltd., Acutest Oy and the Finnish Maintenance Society. References Beattie, A G (1983) Acoustic emission, principles and instrumentation. Journal of Acoustic Emission, Volume 2, Number 1/2, pp. 95-128. Berry, James E (1991) How to track rolling element bearing health with vibration signature analysis. Sound and Vibration, November, pp. 24-35. Brüel & Kjær (B&K)(1984) Instruction Manual. Acoustic Emission Transducers and Preamplifiers. Revision March 1984. 17 p. Li, C James, Li, S Y (1995) Acoustic emission analysis for bearing condition monitoring. Wear 185, pp. 67-74. McFadden, P D, Smith, J D (1984) Acoustic emission transducers for the vibration monitoring of bearings at low speeds. Proceedings of the Institution of Mechanical Engineers, Part C, Vol 198 No 8, pp 127-130. Miller, R K, (Technical Editor), McIntire, P (Editor) (1987) Nondestructive Testing Handbook. Volume 5: Acoustic Emission Testing. American Society for Nondestructive Testing. 603 p. ISBN 0-931403-02-2. Smith, J D (1982) Vibration monitoring of bearings at low speeds. Tribology International, Volume 15, Number 3, June, pp. 139-144. Tan, C C (1990) Application of Acoustic Emission to the Detection of Bearing Failures. Proceeding of the Tribology Conference, Brisbane 3-5 December 1990. The Institution of Engineers Australia, pp. 110-114. Tandon, N and Nakra, B C (1990) Defect Detection in Rolling Element Bearings by Acoustic Emission Method. Journal of Acoustic Emission. Volume 9 Number 1, pp. 25-28.