I recently heard a presentation regarding inaccuracies in data collection. The presenter suggested (among other things) using the actual calibrated sensitivity to reduce errors.

Who out there is following this as a practice? Specific to CSI, are there any other changes required other than in the data collector?

Sensor sensitivity can be set at the point level in DBASE and DBUTLY (as well as the override in the data collector). For our permanently mounted accels, we'll put in the actual number off the certificate.

Patrick,

How about for walk around route-use accels?

All I have is walk around and i don't put in the actual value from the certificate.My sofware says 100mv/g and my actual is 97mv/g..
Thats close enough for my walk around, and I have had no problems at all. If my accel was much lower than 97 i would change it in my software or get another accel.

Mike

For walk-around work I don't enter the actual sensitivity. I use the same sensor on the same points, so absolute accuracy is not a major consideration.

As mentioned, for permanent installs I input the actual sensitivity. And for jobs such as acceptance testing I use an accel that has a cal sheet available and use the actual sensitivity.

Danny, I'll bet your speaker also said to NEVER take data with a hand held sensor, right?

Danny.

Many vib analysts use accelerometers with 10-15% amplitude accuracy. These accelerometers are cheaper than those with 5% amplitude accuracy.

The 10-15% accuracy means that the accelerometers are 100 mv/g plus or minus 10-15%. We configure our databases with a nominal 100 mv/g sensor sensitivity so we don't need to be concerned about the actual sensor sensitivity.

For most of the customers we service, accuracy of +/- 10-15% is acceptable.

If we need less error, we should:
1. Purchase accelerometers with amplitude sensitivity closer to 100 mv/g
2. Calibrate our sensors using a portable shaker and use the sensor sensitivity override in the CSI 2120 and 2130 data transfer programs.

Let me know if I can help.

On some of mine the sens is 1.010. I use that as the sens rather than simply 1 V/g.

On a broad overview - I can't speak for all instrumentation but some may require you to input transducer sens X 0.707 to cause it to read 0-P.

Rusty's approach seems practical for the most part. On permanent installations, you know someone will change out a sensor at some time, but are they going to change the calibration?

One of the criteria in evaluating sensor systems is interchangeability of components. This does not imply that one does not verify or have calibrated the sensor systems.

For axial position, one does calibrate the physical position to the readings. It is not likely that axial probes will be changed in service, and if they are changed in service it is likely to be a mess.

Looking at what API 670 (on protection systems, or is it has to say, one sees that sensitivity should be 100 mV +/- 5% in the testing temperature range (20 - 30 degrees C), but outside this range the tolerance goes up to +/-20%. This standard doesn't allow this much variation for proximity probes (with a larger testing range of 0 to 45 degrees C) and cables.

Frequency response varies over the frequency range. Also, one is not going to put in the mounted frequency response calibration for an accelerometer.

People generally do something similar to Rusty's approach.

Other provisions require 1% accuracy (? not specific) of full scale range.

I had the same belief as Rusty-I used the same equipment at the same point every time, so accuracy was relative. 10% or even 15% wasn't a problem since it was constantly 10% or 15% off.

Now I have someone else collecting data on the same routes and we use a total of 4 different accels, so the sensitivity can affect the repeatability of the readings. I don't necessarily need better accuracy, just more consistant inaccuracy.

Rusty,

No, he never said not to take readings with a hand-held accel, but if anybody there had said they use a 2" stinger, he might have.

Danny, I am using CSI as well for route collection and use 1.01 V/g and 96.n mV/g, etc..

I have route tech's as well. If you have your box set for a specific sens nn.nn mV/g on the equipment within that machine database it will ask for 'transucer' change. So if you had been using the 1.01 on a machine route and change routes where a 97.3 mV/g was needed, the analyzer will ask 'transducer change' to give you the opportunity to change to the correct one. You may want to color code the transducer and route listing and have that route list for your tech in that specific background color. By using a specific # in the database from machine to machine it will prompt you automatically unlike the generic 100 mV/g.

Danny, you could do like I do and buy your accels 10, 20, or even 40 at a time on Ebay and then sort through them and pick out the ones with nearly identical sensitivity. You'd be surprised the bargains you can pick up. For example I just bought 10 new Wilcoxon 793V's for $18 apiece (I like to use them when balancing in 'mils').

If you want I can hook you up with a "matched set". (Of course, that would cost extra obviously)

This message has been edited. Last edited by: rustythevibeguy,

I would buy them like that but there is some guy from Arkansas always bidding things up.

Are Kansas what?

Was I clear regarding the CSI and exact trans sens? The box will ask for correct trans if an exact sens is set and you change transducers or routes so you can always be spot on.

Another thought to consider is that your sensitivity is going to be frequency dependent. How do you incorporate that variability into specifying the sensitivity value into your data collector?

We just use the nominal 0.1 v/g. I figure that for the vast majority of the time, we're not that interested in absolute numbers, but we are looking for changes in spectral patterns etc. But then, we don't do much in the way of acceptance testing and the like.

quote:

Was I clear regarding the CSI and exact trans sens? The box will ask for correct trans if an exact sens is set and you change transducers or routes so you can always be spot on.

Sam, I thought you were perfectly clear, and I also thought Rusty's response (matched sets of accelerometers) was clear.


Ian

Take Rusty’s transducer of choice his $18 793V from Wilcoxon. It is calibrated at 25 degrees C (calibration nominal with +/- 10%). At 80 degrees C (176 degrees F) the calibration want be the same; the datasheet says nomally +3%.

Then there is the frequency response of +/-10% (surely stud mounted with best surface conditions) from3 – 3,500 Hz (+/- 3dB out to 7,000 Hz). The mounted resonance will affect this.

Mechanically, the machine will shake in a direction not inline with the sensor in addition to the inline direction. The sensor has a transverse axis sensitivity stated as 5% of axial (seems odd way to state this). Base strain also affects the sensitivity (0.0005 micro ips/micro-strain).

Then there is electrical noise to be added, seems fairly small on the datasheet. Hopefully, there are no electromagnetic fields present (darn motors/electrical – 25 micro-ips/gauss – a small field will not be significant).

Output impedance is given. With a high iput impedance measurement system this should have a small effect. Did you have an excessive length of cable run back to the instrument?

Not bad for an $18 (and U.S. dollars aren’t what they used to be). All-in-all, maybe we shouldn’t go out the door. It’s scary out there.

What’s the harm in using the nominal sensitivity with all the other effects? Of course if you take your data at 25 degree C environments only, and the vibration frequency components lie in the sweet spot (or is that sweat spot) of the mounted frequency response curve, etc. maybe you do get something by using the given value for sensitivity instead of the nominal one (leaving out human errors for using the wrong sensor/calibration factory and complicating the process).

If one wants to do this correctly, I suppose with some effort one could add error bars around the measurements to show the client regions where the vibration levels most probably lie (confidence intervals for those fond of statistics). ... but is this needed for condition monitoring?

Danny,
I think the key to routine monitoring is repeatability. We are very cafeful in ensuring that the same Accel and Magnet is used every time, in 5 years of intensive monitoring, I have never had to change the Accel once, we have conducted simple comparisions between Accels with the same cable and data collector, and noted very little change from one Accel to he other, certainly not enough to cause any alarm. If you use the same Accel and magnet, all the time, it really doesn't matter what the sensitivity is, as the data will always be repeatable. Accels are very robust, and if looked after last a very long time, we have some permanent installations that have been in for over 10 years, we have more trouble with the cables then the Accels!

And finally, when it all goes to hell, it doesn't matter what the sensitivity is... data errors of 1, 5, 10, 15, or even 20% (or more) won't matter because the defect/fault/impending failure will be hard to miss!

Rusty,

If all I wanted to accomplish was preventing bearing seizures, I don't really need a data collector to do that. Where it matters is in the trend. Some situations exist that make it so that sometimes, we cannot perform an exact duplicate of each route. That is to say, we cannot be sure of taking the same readings with the same accels every time.

Bill,

You seem to suggest that because there are many sources for errors, we should abandon attempts to minimize them. Like you said, some are within our control and others are not. I am not attempting to be able to say with absolute accuracy that the amplitude is 5 g's p-p but to be able to say with a higher degree of accuracy that the amplitude has increased or decreased. If I use (and do not correct for) an accel one time that has a sensitivity of 89.7 mv/g and the next time one that has 104.5 mv/g, I have induced errors that could have been compensated for along with the others that can't be compensated for.

I think the best we can do should be, do the best we can! Don't add error from what you can control to what you can't control.

I like the approach of trying to go for the greatest accuracy with spot-on sens setting with proper resolution for the machine with the correct cable. Yes, my tech's complain about switching to micro-dot cables an such but!!! pays the same.

If you set the box generically you'll incorporate an unknown error. Why, when you can take a transducer specific to that job with its exact calibration sens. It's very easy with a route as the transducer's sens it already set and it will prompt for the correct transducer to match the machine on the route.

It can be important on the trend. Afterall, we are not just trending vibration. We are looking for change; that change may be from a poor PM function schedule that needs altering. Is the trend perdiodically changing on a very repeatable basis? May not be a fault from the machine, so, our accuracy must be able to detect this or at least accurate enough that we have confidence in our data.