Pumps & Systems, April 2007
Sometimes standard methods of vibration analysis are not enough to provide a true picture of the health of a pumping system, especially in the early recognition of an under-lubricated machine component.
Two advanced technologies, analytical methodology and autocorrelation, can be helpful in this regard. When employed together, they can be very effective for monitoring and evaluating the state of lubrication in rotating machinery.
Analytical methodology (see Reference 1) provides a means to consistently examine the information available in the higher frequency (greater than 1-kHz or so) range of signals from an accelerometer placed on or near a bearing housing. This analysis actually measures "stress wave" activity in a metallic component. Such stress waves are associated with impact, friction, fatigue cracking, etc. that eventually generate faults in machine components, such as rolling element bearings in motors and pumps.
When a rolling element impacts a defect on a bearing raceway, stress waves propagate away from the location of the defect in numerous directions. This introduces a short duration ripple on the machine surface that generates an output signal from an accelerometer mounted on the surface. Stress wave analysis is a powerful complementary tool that can detect a range of faults and problem situations that routine vibration analysis alone might miss.
Autocorrelation (see Reference 2) provides an alternative to spectral analysis of captured vibration waveforms. This methodology exhibits specific strengths relative to spectral analysis, including: 1) the ability to identify the presence of low frequency periodic activity in the waveform, and 2) the ability to separate out non-periodic (random) activity.
When a rolling element bearing in a motor or pump lacks sufficient lubrication, the output signal from an accelerometer monitoring that bearing can be expected to exhibit a significant increase in its high frequency sector (greater than 5-kHz) with little or no periodic recurrence, except in special cases. Since the energy associated with friction lies above the 5-kHz range, it is recommended the accelerometer (sensor) used in the Route data collection be mounted with a rare earth flat magnet on a smooth flat surface (such as a mounting pad) free of paint and debris. A significant increase in the g‑level of the waveform, combined with a low value in the autocorrelation waveform, may indicate excessive friction because the component under study is not being adequately lubricated.
The following two cases demonstrate the application (or non-application) of these technologies in real world situations. In the first case, vibration data was collected on a bearing in a motor driving a green liquor pump in a paper mill. Despite significant peaking evidence, no action was taken and the result was disastrous. In the second case involving a bearing on a scrubber induction fan, action was taken immediately to restore lubrication, enabling the machine to continue operating without downtime.
Case Study #1
During a periodic route-based vibration data collection, a loud audible noise in the technician's headphones triggered the acquisition of a data set. The point of origin was on the inboard side of a pump moving green liquor in the pulping operation of a paper mill. The data (spectra and wave form) are presented in Figure 1, showing a peak g‑level of around 13, which exceeded the recommended fault level of 12 for a pump operating at that speed. The spectral data implied a low frequency periodic response (as might be caused by a cage fault).
Figure 1. Data acquired on the pump inboard February 24, 2006. Sensor is an accelerometer.
The autocorrelation data presented in Figure 2 clearly show the activity to be random, with no recurring low frequency activity. This signature, high g‑level and little to no periodic activity strongly suggest friction due to lack of lubrication.
Figure 2. Autocorrelation data computed from the waveform shown in Figure 1.
Since the g‑level was only slightly greater than the recommended fault level, no corrective action was taken at the time, and the pump experienced catastrophic failure before the next scheduled (monthly) collection of vibration data. The cause of the costly failure was later determined to be "lack of lubrication."
Case Study #2
A loud noise in the technician's earphones driven by a sensor located on the outboard bearing housing of a scrubber ID fan triggered the collection of data at 14:02 EST. The waveform had a peak g‑level of 13, and it was totally random, indicating no periodic recurrence in the spectral and the autocorrelation data. This fan is lubricated by an oiler.
When the technician lifted the sight glass bowl, slightly permitting a little extra oil to enter the oil reservoir, the noise in the headphone immediately subsided. A set of data taken three minutes later showed the peak g‑level reduced to 1.6, well below the recommended alert level of 6 for this particular application.
Summary
Excessive friction in bearings on rotating machinery, generally induced by a lack of sufficient lubrication, leads to the emission of short duration stress waves (in the spectral domain, the energy from stress waves will exceed 5-kHz). An accelerometer, properly mounted (such as a flat rare earth magnet on a flat, smooth surface free of debris and paint), will provide reliable detection of the friction-induced stress wave activity.
The methodology above, coupled with autocorrelation analysis, proves to be an effective tool for the detection and identification of friction-induced stress wave activity. With timely detection and fast corrective action, the working lives of machines being monitored can be extended.
References
-
Jim Crowe and Jim Robinson, "An Overview of the PeakVue Methodology Demonstrated with Case Studies," 30th Annual Meeting Proceedings of Vibration Institute, Louisville, KY, pp. 149-158.
-
James C. Robinson, "Autocorrelation as a Diagnostic Tool," 30th Annual Meeting Proceedings of Vibration Institute, Louisville, KY, pp. 79-92.