Go
New
Find
Notify
Tools
Reply
  
-star Rating Rate It!  Login/Join 
Posted
How are small speed changes prevented from causing false alerts when creating a reference constant percentage bandwidth spectrum?
 
Posts: 17 | Location: Cyprus | Registered: 22 September 2006Reply With QuoteEdit or Delete MessageReport This Post
Posted Hide Post
Please see the attached also here.
Regards
Wayne

PDF DocCPB_Measurements_-_early_fault_detection_with_minimal_risk_of_false_alarms.pdf (203 Kb, 28 downloads)
 
Posts: 17 | Location: Cyprus | Registered: 22 September 2006Reply With QuoteEdit or Delete MessageReport This Post
Posted Hide Post
How does this sound then?

- As low energy signals will be buried in the more powerful vibration signals, incipient ball bearing faults or incipient gearbox faults will not be detected. At least they will not be detected before they generate a vibration signal higher than the low frequency domain signals
- If only broadband measurements are carried out (e.g. 10 -1000Hz according to ISO or VDI, or with another selected bandwidth) no changes will be detected before the increasing component will reach the level of the highest component in the measured band
- If spectrum comparisons are performed, the changes will be detected as soon as one band in the spectrum increases more than 3-6dB, thereby giving a very early warning
- Having concluded that spectrum comparison is needed to obtain the earliest possible warning of a fault developing in a machine, one may be tempted to use the storage and comparison capability of most single channel analysers
- Very small changes in speed will shift the position of the peaks and result in large differences and give false exceedances, and therefore false warnings
- The solution is to perform a kind of data reduction, i.e. group the lines in the FFT spectrum together in larger groups, thus allowing changes inside each of these groups without interfering with the overall level of these groups. The new spectrum created in this way is referred to as a synthesized spectrum. Of course, the frequency bands in this synthesized spectrum must be wide enough to absorb the random changes in the signal, but narrow enough to be able to detect small changes in the major component of the spectrum
- So this addresses the matter of small changes; which logically leads on to significant speed changes. If a machine changes operation speed more than that allowed keeping the frequency components within one bandwidth, the detection system described up to now will also give false warnings. If the spectrum is synthesized in such a way that we obtain a Constant Percentage Bandwidth spectrum, the effect of a speed change will be the same all over the spectrum
- If the speed of a gearbox increases 15% the tooth meshing frequency and all harmonics originating from the gearbox will also increase 15% in frequency. Thus we can compare the new synthesized spectrum with our reference mask by shifting the reference mask 15% up in the frequency domain. All fundamental frequencies, harmonics and sidebands, etc., will line up again, and false warnings will be avoided
- Therefore, when using spectrum comparison for fault detection, there will always be a risk of having false warning signals. If the resolution in the synthesized spectrum is high enough, the majority of these false warnings can be identified immediately and disregarded
 
Posts: 17 | Location: Cyprus | Registered: 22 September 2006Reply With QuoteEdit or Delete MessageReport This Post
Posted Hide Post
Here is the study on the broadband spectrum: http://caltecheerl.library.caltech.edu/261/00/DYNL9701.pdf
 
Posts: 2525 | Location: Borneo | Registered: 13 February 2005Reply With QuoteEdit or Delete MessageReport This Post
 Previous Topic | Next Topic powered by eve community  
 


Copyright © 2004-2008 NetexpressUSA Inc. All rights reserved.