Sign up Calendar Latest Topics Donate
 
 
 


Reply
  Author   Comment  
Shurafa2

Sr. Member
Registered:
Posts: 170
Reply with quote  #1 
Hello forum,

I've tried to verify my basic understanding of certain terms by referring to recognized standards/
references like:

1. ISO 13372 (Condition monitoring and diagnostics of machines — Vocabulary) and 
2. API 670 (Machinery Protection Systems)

 

For noise, I did not see it listed as a stand-alone term in the definitions. 

Fore noise floor, API 670 did not define but it provided a maximum acceptable limit.

Below is the text from ISO 13372.
6 Data characteristics
6.3 noise floor: level of noise present in a system with no exciting signal present

Do you think this definition is adequate and it reflects what you use in the industry?

Regards- Ali M. Al-Shurafa
electricpete

Sr. Member
Registered:
Posts: 647
Reply with quote  #2 
Yeah, it's a tough one  Typically we talk about raised noise floor.  And roughly it means the spectrum in between the discrete peaks is raised due to random / broadband excitation.  One thing we know is that for a given signal, the appearance, level of this noise floor depends on the bin width (lower bin width gives lower noise floor) and the averaging (averaging tends to make the noise floor smoother).    PSD is better suited to evaluation of random signals than normal FFT spectra in the sense that the level of the floor of PSD does not depnd on the bin width.   With regular spectra, for the most  we judge noise floor by comparing to history or similar machines or past history of this maching, making sure that these comparisons have the same data collection parameters.   

Sorry, that doesn't really answer the question, just some rambling...
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #3 
Yeah, leave it to me & ePete to always jump on these terminology threads...

Our glossary gives:

Noise [Signal] – The total of all interferences in a measurement system that are not present at the start of the measurement train. This includes all effects between the Transducer until Digitization. See 10.101.001 (Sensor Fundamentals).

To me, noise in the general sense is any undesired and confounding component of a signal. Often, it is an artifact of the measurement chain and is not part of the physical measurand. Many only use noise to apply to dynamic signals (where f(t)), but in my definition, the systematic and stochastic error of a static signal can be considered noise.

Noise is not just broadband noise. AC ground loops giving you odd harmonics of LF is also clearly noise.

This definition only considers electrical noise to be noise, but one COULD also consider real peaks in the spectra noise if they come from machines not in the scope of the measurement. For example, monotonal vibration from an HVAC compressor is noise if we are looking at the undesired effects on nearby microscopes.

The entry for our group's glossary for noise floor is:

Noise Floor – Sometimes Noise Carpet. The signal produced from the sum of all available Noise sources. Spectral Peaks below the Noise Floor will not be resolved.

This infers that all noise sources form the noise floor, event he harmonic ones. I think many will only consider the broadband noise to contribute though. This is clear in the quantification of noise floor that most dynamic sensors give. Most accelerometers give a measure of μg/√Hz, often at different frequencies. This parameter is referred to as the Noise Density, Equivalent Noise Floor and Spectral Noise.
Shurafa2

Sr. Member
Registered:
Posts: 170
Reply with quote  #4 
One good method to examine how good a definition is is to compare it to common examples of the term. Some of the definitions mentioned above for noise and noise floor (or floor noise) are not really good enough, IMO.

In my small world, noise can come from:

1. the instrument loop (e.g. loose wires, unstable sensor)
2. the environment of the instrument loop (e.g. interference with power, other probes)
3. the equipment (mechanical (e.g. intermittent sudden moves due to looseness), electrical (e.g. 2LF from the design), process (e.g. turbulence, cavitation))
4. the environment of the equipment (e.g. beating with sister machine)

Sometimes, the noise is a problem to the measurement (not necessarily for the monitored equipment). Sometimes the noise is the problem itself that we need to resolve to protect the equipment from damage.

Now, is what we see in the signal a noise or not (in terms of magnitude)? API 670 states the thresholds as below:

"5.2.1.12 The accelerometer transducer shall have a noise floor no higher than 0.004 g rms over the frequency range specified in Table 1."
"5.2.3.13 The velocity sensor shall have a noise floor no higher than 0.004 mm/s rms over the frequency range specified in Table 1."

I did not see the criteria for relative probes. To me these specifications might be suitable to qualify instrumentation (for procurement) but they would not be the practical limits that would trigger my attention in the field to diagnose anything (including a false signal).

Perhaps my judgment is not what others would use in their normal jobs.

Regards- Ali M. Al-Shurafa
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #5 
Quote:
Originally Posted by Shurafa2


Now, is what we see in the signal a noise or not (in terms of magnitude)? API 670 states the thresholds as below:

"5.2.1.12 The accelerometer transducer shall have a noise floor no higher than 0.004 g rms over the frequency range specified in Table 1."
"5.2.3.13 The velocity sensor shall have a noise floor no higher than 0.004 mm/s rms over the frequency range specified in Table 1."



It is very important that this is only for qualification of the sensors. This parameter of a sensor is a quantification of the inherent static (thermal noise or Johnson–Nyquist noise) to that sensor. It says nothing about the practical levels of signal.

Also, it can be difficult to separate noise sources 1+2 from 3+4, but it is definitely necessary. By default, when someone mentions signal noise, I think only of sources 1+2 (electrical noise?). Sources 3+4 I usually don't refer to collectively at all, much less as noise, though I could call it physical noise.
electricpete

Sr. Member
Registered:
Posts: 647
Reply with quote  #6 
In the EE world of signals analysis, there is no doubt that "noise" is associated with unwanted signals which may obscure the signal that we are interested in. 

But I don't think that is the meaning commonly embraced in the colloquial vibration world among people who spend their days doing routes when referring to "raised noise floor".  In that world, a "raised noise floor" typically means  any broadband like signal, including those originating from your machine that you should be interested in.   For example people talk about "raised noise floor" occurs in stage 4 of rolling bearing defect degradation or raised noise floor in a certain part of the spectrum for rub, cavitration, turbulence etc.    I have seen that terminology used in training materials, on the forums, by the people I work with.  I'm pretty sure that phrase is used on Technical Associates charts.   And that is the terminology I use myself.  (Personally I learned most of what I know about vibration and vib terminology on the forums starting with reliability-magazine.com... learned about signal processing in college long before that, and I tend to treat them as two different worlds of terminology).

I think I can see where that terminology might have arisen.   Noise is used like an adjective to clarify what floor we're talking about (the floor on the bottom of the spectrum).  If there was a machine with no broadband excitation, then whatever spectral floor exists is in a sense noise (even if it's leakage, that's not really the part of the signal we're interested in).    When that floor at the bottom of the spectrum raises, we call it a raised noise floor,  even though the source of the rising is usually not noise (unwanted signal).    All of that origin thought process about why we someone called it a noise floor to begin with is lost, and  we just include the word noise in our phrase "raised noise floor" to help people recognize what floor we're talking about. 

I'm not saying that's right, but that's the way I've heard it used so often.     To be more in line with the signal analysis terminology, it would be better to drop the word "noise" altogether and just say "raised floor" or "raised spectral floor" when we're referring to broadband-like signals presumably arising from the machine that having nothing to do with what would be called "noise" in the EE field of signal analysis.  

But if I was trying to teach someone anything about interpretting that spectral floor, the first thing I'd want them to understand is that  the level/magnitude of the floor in both normal and abnormal conditions depends heavily on the data collection setup (bin width, averages).  That is something they might not tell you on a diagnostic chart that shows a raised noise floor as a symptom.
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #7 
Thanks Pete, 100% agreed.
electricpete

Sr. Member
Registered:
Posts: 647
Reply with quote  #8 
Quote:
"5.2.1.12 The accelerometer transducer shall have a noise floor no higher than 0.004 g rms over the frequency range specified in Table 1."
"5.2.3.13 The velocity sensor shall have a noise floor no higher than 0.004 mm/s rms over the frequency range specified in Table 1."


Do they specify spectral setup?  To me it seems nonsensical to specify a noise floor with units g or mm/s without specifying a bin width.   

On the other hand, the noise floor of a sensor could reasonably be specified as a power spectral density, but for PSD the units would have something like hz or sqrt(hz) included in the denominator. 

Or maybe in this case they are referring to an overall / rms calculation over a specified frequency range? That would make sense as a parameter to characterize the inherent noise, but why would you call that a "noise floor"? (it's nothing like a spectral floor). 

Sensors and sensor selection is not a strong point for me.  (Our vibration engineer does that, I'm just a consumer of the data in my role of monitoring/managing motors). 
Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #9 
Pete,

Yes, those values are broadband overalls over the nominal frequency range of the sensor.

I had to do a bit of transducer speccing for an IOT sensor development. The noise floor has been the main disadvantage of MEMS (VC) accelerometers for the past 20 years, but there are a few sensors released in the last few years that have really made strides in decreasing the noise floor. Its almost to the point where things like peakvue will be accomplished with MEMS sensors, which would be a game changer.
Shurafa2

Sr. Member
Registered:
Posts: 170
Reply with quote  #10 
Quote:
Originally Posted by electricpete


Do they specify spectral setup?  To me it seems nonsensical to specify a noise floor with units g or mm/s without specifying a bin width.   

[....] 

Or maybe in this case they are referring to an overall / rms calculation over a specified frequency range? That would make sense as a parameter to characterize the inherent noise, but why would you call that a "noise floor"? (it's nothing like a spectral floor). 



Pete,

Below is API 670 Table 1 of Machinery Protection System Accuracy Requirements which is referenced to in the previous posts. The "Accuracy Requirements" are like linearity verifications and maximum deviations based on ambient conditions.

I do not think, for the hardware of protection systems, the vendor will be concerned with the bin width. At this level, there is no FFT process. But maybe I'm missing something.

Regards- Ali M. Al-Shurafa
API 670 Table 1 Machinery Protection System Accuracy Requirements.png

Curran919

Avatar / Picture

Sr. Member / Supporter
Registered:
Posts: 599
Reply with quote  #11 
Ali,

The sensor's amplitude linearity there has nothing to do with the sensor noise floor.
Shurafa2

Sr. Member
Registered:
Posts: 170
Reply with quote  #12 
Curran,

Perhaps you are right, linearity has nothing to do with the sensor noise floor. Showing the table is a response to Pete's comment to see the rest of the setup used to define the max noise.

Regards- Ali M. Al-Shurafa
John from PA

Sr. Member
Registered:
Posts: 963
Reply with quote  #13 
Quote:
Originally Posted by Shurafa2


Pete,

Below is API 670 Table 1 of Machinery Protection System Accuracy Requirements which is referenced to in the previous posts. The "Accuracy Requirements" are like linearity verifications and maximum deviations based on ambient conditions.

I do not think, for the hardware of protection systems, the vendor will be concerned with the bin width. At this level, there is no FFT process. But maybe I'm missing something.

Regards- Ali M. Al-Shurafa


In the definitions section, 3.1.158 resolution discusses "bins" but not in much depth.  Also Annex N has some discussion on FFT but again not much depth.

Keep in mind that AI 670 was not intended to be an instruction manual nor replace knowledge.  
electricpete

Sr. Member
Registered:
Posts: 647
Reply with quote  #14 
Curran's explanation "those values are broadband overalls over the nominal frequency range of the sensor." clears things up for me.   

That is a reasonable measure of sensor/circuit inherent noise, and it will not depend on bin width (unlike spectral floor which will always depend on bin width). 

The words about noise floor in sections 5.2.1.12 and 5.2.1.13 quoted by Ali are "over the frequency range specified...".  I initially interpretted that incorrectly as a maximum value of the spectral floor within a frequency range, but now I see it makes more sense it would be an overall over a frequency range. 

The phrase "noise floor" might be a reasonable way to describe this sensor/circuit inherent noise, but the confusion for me arose because many of us also commonly use the phrase "noise floor" to refer to "spectral floor', which is something completely different. To illustrate my assertion that this is common usage (noise floor = spectral floor), look at comments two days ago from respected member George D here.    (I'm not picking on you George, many people use that terminology).   

I apologize for bringing confusion into this thread by initially interpretting "noise floor" to mean "spectral floor".     If we have to use the term noise floor, I suggest we only use it only in reference to the sensor/circuit inherent noise overall and not to the spectral floor. 
Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.