SUBMITTED BY: Dustin Morris
TITLE: Program Manager
EQUIPMENT: E4204 2 Fiber Predryer
TECHNOLOGIES APPLIED:
Vibration and Ultrasound
PROBLEM DESCRIPTION:
Working in a predictive maintenance program at a site with over 3000 rotating assets for 9 years my fellow technicians and myself had become confident in making calls on “standard machinery”; i.e. pumps and fans with a rotating speed of 1100rpm and above. The site has employed a predictive maintenance program for over a decade and has seen it grow from two full time employees, performing vibration tasks, to fifteen full time employees performing vibration, ultrasound, infrared, motor current analysis, and oil analysis tasks. However given the sites size and years of operation, over twenty, a majority of the plants slow speed equipment was beginning to suffer, slow speed being defined as equipment with a rotating speed of less than 500rpm. Due to a string of recent failures the question was posed: How do we properly monitor the condition of bearings rotating at slow speed?
Over the past year our vibration program had been experimenting with various methods of monitoring the condition of slow speed bearings. Several methods of collection with an Entek DataPAC 1500 were explored. This included connecting an UE 9000 ultrasonic gun to the DataPAC and collecting spike energy and time waveform data. This provided nice looking data but was cumbersome for the technician to perform on monthly routes. Especially considering a majority of the slow speed points were outside and only accessible with ladders. A comparison of data with a shear compression 500 mV/g accelerometer and the same collection specifications we used with the UE9000 provided equally acceptable results. This also proved to be more economical. At this point there was a lot of experimenting with the actual collection specifications. We needed to come up with an acceptable fmax to get the correct number of revolutions; yet still provide good resolution and an acceptable time for collection. It also needed to be simple so it was easy to apply to machines with speeds ranging from 6rpm to 500rpm. For these reasons we decided to use an orders based collection. The slow speed collection specification developed is as follows:
Collection Specification for Velocity Spectra |
|||
| Transducer: | LF Accelerometer | Signal Detection: | Peak |
| Window: | Hanning | Number of Averages: | 2 |
| Maximum Frequency: | 160 Orders | Type: | Linear |
| Number of Lines: | 800 | Percent Overlap: | 50 |
| Filter: | None | ||
Collection Specification for Time Waveform |
|||
| Transducer: | LF Accelerometer | Signal Detection: | Peak-Peak |
| Window: | Rectangular | Number of Averages: | 1 |
| Maximum Frequency: | 160 Orders | Type: | Linear |
| Number of Lines: | 1600 | Percent Overlap: | 0 |
| Filter: | Smart HP | ||
Collection Specification for gSE Spectra |
|||
| Transducer: | LF Accelerometer | Signal Detection: | Peak-Peak |
| Window: | Hanning | Number of Averages: | 2 |
| Maximum Frequency: | 160 Orders | Type: | Linear |
| Number of Lines: | 800 | Percent Overlap: | 50 |
| Filter: | 500Hz gSE | ||
INVESTIGATION:
Using the collection specification above provides a varying bin width depending on speed but seems to be acceptable across the speed ranges. For example a machine rotating at 6rpm results in an fmax of 960cpm. With a hanning window applied this results in a bin width of 1.8cpm(960/800 X 1.5). At the other end of the range a machine rotating at 500 rpm results in an fmax of 80Kcpm and a bin width of 150cpm. This appeared to be adequate in identifying the fundamental frequency of most bearings. The only problem with this collection specification is that a bearing rotating at 6cpm results in a collection time of 5 minutes and 7 seconds.
The following case study, while not providing all of the answers, did go a long way in enlightening us as to what may be most effective. In November 2005 a relatively new vibration technician came to me requesting assistance analyzing vibration data from a trunion bearing on a hot air rotary dryer. It was data taken with the new collection specifications (see above) that the technician recognized what he thought to be a problem. An I.M.I. 626A02 500mV/g accelerometer and a DataPAC 1500 were used to collect the data.
Figure 1
Figure 2
The bearing position in question is on the ring gear side of the dryer. The velocity spectrum (upper left corner of Figure 1) for this position was very complex with a raised noise floor from 250cpm to 1500cpm. Many of the peaks within the raised noise floor were spaced at the running speed of the trunion roll. The time waveform (bottom of Figure 1) also exhibited some impacting at 1xRPM of the trunion roll. The spike energy data (upper right corner of Figure 1) was the most interesting. It contained a harmonic family with peaks spaced at 125cpm. At this point in time the fault frequencies were still not entered. Early on in diagnosis there was some speculation that the vibration was from the 160.5” tire contacting the trunion roll or from the meshing of the pinion and the ring gear. This was quickly discredited as the technician pointed out that he had the same vibration signature on the opposite side of the dryer on a different trunion bearing. This bearing is over 31 feet away. Also the bearing on the opposite side of the trunion in question had no such vibration signature. Given that we clearly had a harmonic family with a fundamental frequency of 125cpm in our spike energy data, we decided that it was time to research fault frequencies.
The equipment file contained a very informative repair and instruction manual. From it we determined that we were dealing with a Davenport 12’ x 60’ rotary hot air dryer. A 50HP 326T 1760rpm Toshiba motor drives a Falk gearbox, model number 405A3, with a rated output speed of 30rpm. The output shaft of the gearbox is coupled to a pinion gear with 27 teeth driving a ring gear with 250 teeth. On each end of the dryer shell are two forged steel tires with an outer diameter of 160.5”. These ride on two trunion rolls with an outer diameter of 36” each. The bearings on each side of the trunion rolls are SKF 22230CC.
Using the information from the equipment file, fault frequencies were entered into the vibration database. This revealed that an SKF 22230C bearing with an inner race rotating at 15 rpm produced a ball pass outer race frequency of 123cpm. This happens to fall within 2cpm of the harmonic family identified in our spike energy data. Armed with this information the technician entered a work order to change the bearings at position 16 and position 18 (see Figure 2).
Fault frequency of ball pass outer race with harmonic cursor.
Figure 3
On April 13, 2006 site mechanics began the job of changing the trunion bearings on positions 15 and 16. It was decided by site maintenance coordinators to complete the work on positions 16 and 18 at different times due to the amount of down time involved. Upon completion of the job on positions 15 and 16 follow up readings were taken and the defective bearing was retrieved for analysis.
Figure 4
Figure 5 Spike Energy Spectra Data taken before repairs is on the left; data taken after is on the right.
The damage shown to the bearing from position 16 is indicative of fatigue wear,see Figure 4 above.
These bearings have turned nonstop for the last 14 years except for planned maintenance.
It is obvious from the photograph in Figure 4 that the outer race of this bearing did have multiple defects. The spike energy spectra from Figure 5 also show that after the repair the ball pass outer race defect frequency was no longer present. Unfortunately the velocity spectrum (Figure 6) and the time waveform data (Figure 7) show little change.
Figure 6 Velocity spectra Data taken before repairs is on the left; data taken after is on the right 
Figure 7 Time Waveform Data taken before repairs is on the left; data taken after is on the right.
In this case the spike energy data proved to be the most useful. This also appears to be the case with position 18. The data in Figure 8, from position 18, is very similar to the data in Figure 1, from position 16. The spike energy data from position 18 shows that this bearing also has an outer race defect.
Figure 8 Data collected on position 18.
Knowing that this bearing is experiencing the same failure as position 16 gave us a wonderful opportunity to experiment with the different technologies and their effectiveness to determine the condition of a slow speed bearing. One (Position 16) that has been newly installed and shows no signs of a defect and one (Position 18) that has been in use for several years and is showing signs of a ball pass outer race defect. Both bearings are SKF 22230C bearings with an inner race rotating at 15rpm on the same machine. They are both in a similar environment and are exposed to similar operating variables.
The first technology explored was thermography. A Mikron 7515 IR camera was used to take images of the thermal energy of both bearings. Since this equipment is outside the images were taken early in the morning to assure that the sun was not giving a false reading. This had been experienced on the day prior due to the angle of the sun in afternoon.
Figure 9 The temperature on position 16 is 88.3°F.
Figure 10 The temperature on position 18 is 89.4°F.
The temperature difference between position 16 and 18 was only 1.1°F. This was hardly the smoking gun we were hoping for. A temperature difference this slight does not seem good enough to make a reliable call on bearing condition.
The next technology explored was ultrasound. We have access to an UE Ultraprobe 2000, UE Ultraprobe 9000, and UE Grease Caddy. Because access to the bearings could not be achieved safely a magnet-mounted accelerometer was used to collect the readings. With the UE Ultraprobe 2000 and UE Ultraprobe 9000 set at 28kHz all three instruments produced similar results. Since this mounting method was used none of the instruments registered more than 1 dB of energy. The ability to hear an audible signal through the UE instruments in the field was virtually impossible, due to the low energy level, mounting method, and the background noise. Despite this setback we decided to make comparative sound recordings to listen to and analyze in an environment where the background noise could be controlled. Using the UE Grease Caddy and an iRiver iFP-790 mp3 player/recorder two one minute sound files were recorded, one on the repaired bearing on position 16 and one on the defective bearing on position 18. The files were then downloaded to a computer and converted to an mp3 format. The mp3s were then imported into software known as Audacity. Audacity is software used to view the sound wave of an audio file and make enhancements to it. In this instance it was only used to listen to and view the files. Despite not being able to hear anything in the field once imported into Audacity the resulting sound waveforms showed a marked difference. Using Audacity, WAV files were created to import into UE’s Spectralyzer software. The UE Spectralyzer software produced an identical sound waveform, however due to our inexperience with the software an acceptable spectrum was not produced.
Screenshot from Audacity software.
Figure 11 The sound wave on top is from position 18, the one on the bottom is from position 16.
In this case trying to make a call in the field with ultrasound would prove extremely difficult; however recording a sound wave for analysis in a controlled environment does provide some useful information. It does not pinpoint an exact defect frequency but it clearly shows the difference between a healthy bearing and a defective bearing.
Next up was coupling an UE Ultraprobe 9000 to a DataPAC 1500. This method did not provide us with a velocity spectrum but it did provide us with an acceptable spike energy spectrum (Figure 12). We were also able to collect time waveform data; this data had impacting at 12cpm which none of the fault frequencies can explain. For that reason the time waveform data does not seem as useful as the spike energy data at this time. Further experimentation with collection specification and methodology could be performed to provide better time waveform data.
Figure 12 Spike energy data collected with ultrasound and vibration data collector.
We realize that the aforementioned technologies and equipment may not be the only items available on the market to monitor the condition of slow speed bearings. However it was the only equipment and technologies that we had available to us at the time this was written.
CONCLUSION:
In summary the two technologies showing the most promise were vibration and ultrasound. Given the data presented here the defective bearing makes its presence known with both technologies. It is the fact that we were able to pinpoint an exact defect frequency and confirm it upon repair that gives vibration the edge. Ultrasound shows promise and we will be experimenting with it further, it did however have some limitations. Attempting to listen to and analyze a defect in the field was extremely difficult. This is due to the small amount of energy from the defect and the competing background noise. Recording the sound waves and analyzing them in a more suitable environment does appear to be an acceptable method of determining a slow speed bearings overall condition. This does require some extra equipment i.e. an mp3 recorder, cabling for the recorder and ultrasound instrument, and software. The vibration collection specification and vibration equipment used to collect the vibration data is available to most predictive maintenance programs employing vibration. Enveloping and other forms of high frequency detection could be used in place of Entek’s spike energy. This would make using specially developed collection specifications for vibration one of the most effective and easiest technologies to apply on a slow speed bearing.
