This is my first post although I have been reading them for awhile.
I am running with 6400 LOR and am wondering if that is to many. I have been taking spectrum for a number of years but am new to time waveform.
Does it make a difference what the F-max is?

PG,

Maybe.

There are a number of things that could affect your results and to really answer, we need to know all of your current settings, including what type of equipment you are using, input devices, collection parameters, etc.

If you can post more information, I'm sure you will get plently of help, but if we do it without all the information, we might be wasting some time.

Wlecome and Good Luck

PG, like Danny said it depends on what your trying to accomplish. In general... For a given LOR, the lower the resolutioin will be as you increase F-max. The higher the LOR for a given F-max, the higher the resolution. The higher the resolution the longer it will take to capture the data and the larger the data file size(e.g., more storage space on your hard drive).

Thanks

It is an engine driven blower with the final RPM's of 5000.
The pilow block bearing is a SKF 22217 and runs at 1000 RPM's. It is driven by a belt and has 190 teeth on the gear that drives the output shaft.
The blower wheel bearing is a SKF 313 and runs at 5000 RPM's. It has 16 vanes and 38 teeth.
Sorry I am new to posting so if it takes me awhile thanks for taking the time to deal with me.

PG,

I wasn't very clear when I said what type of equipment, but we also need to know about your data collection equipment and set-up to give you an answer.

I use a Recip-trap from Dynalco. It is mostly an engine and compressor HP and vibration tool. It has the ability to do spectrum and have been taking data on rotating machines for some time. I work for a nat. gas trans. company and have mostly internal combustion engines. The set up on the pillow block is in g's with a F-max of 12000 and 6400 LOR. The set up on the blower is 60000 F-max and 6400 LOR.
Woould the attachment look right or am I way off base?

Pillow_block_TWF.doc (58 Kb, 53 downloads)

PG,

T (length of TWF)=lor/fmax
T=1/bw
t=sample size/sample rate
sample size=2.56 * lor
sample rate=2.56 x fmax

These are limited by the data collector to something like 4096 or 8192 samples.

Play around with these formulae and how they are related and you will probably find your answer.

Keep in mind that with CSI equipment (and possibly others) you could also be collecting a special time waveform with settings that are different from your settings for the frequency domain.

Without regard to the limitations of your data collector:

T=6400/200 Hz=32 sec=.533 min. * 1000 rpm=533 shaft rotations per time waveform or average.

T=6400/1000 Hz=6.4 sec=.107 min *5000 rpm=533 shaft rotations.

If this is derived from the spectrum collection settings, there will also be a number of averages and possibly a percent of average overlapping that affect the total time.

Generally speaking, with antifriction rolling element bearings you need to see 8 to 10 shaft rotations. Bearing wear will start to show up well beyond your fmax setting, but you are getting a very high resolution look at the low frequencies where shaft related problems will occur.

I would try collecting data with an fmax of 1000 hz and 1600 lor for the 1000 rpm stuff and 5000 hz and 6400 lor for the high speed stuff. There may be limitations defined by your instrument and some variation to these settings is fine, but you Add it to your data collection and continue to collect what you have been for at least 3 sets of readings so that you can track trends in both for a while.

Good Luck,

Danny

Thanks for your help. I will try that. That is what I am after. I don't think there is a problem but want to get data to trend.

I'm now more confused than ever. I think that my software isn't figuring the run orders right. This is what I came up with and there is a good chance I am wrong.
T= 1600/60,000= .027 X 1000 rpm = 27 rotations
T= 1600/120,000 = .013 X 1000 rpm = 13 rotations

T= 6400/300,000= .021 X 5000 rpm = 106 rotations
T= 3200/300,000= .011 X 5000 rpm = 53 rotations
T= 3200/600,000= .005 X 5000 rpm = 27 rotations
T= 3200/1,200,000= .0027 X 5000 = 13 rotations

If I am such a dumby that I can't even get this right I will wait until I can get some formal training. If it is the software I will contact the vendor.

thanks for your help.

Danny's equations are a good start, but there are some additional complications....

The length of the time record depends on the lines of resolution and the Fmax settings (some systems may do this differently, but the waveform captured is usually the sample used to compute the FFT). A higher Fmax frequency gives a shorter time record.

IF you were looking at something like a once per rev impact with a real time analyzer, at a relatively high Fmax, you might see a burst of energy that fills the screen with each impact. It's pretty easy to see impact events.

If you see an impact, the next question is what is the repetition rate between impacts? Once per rotation? More often? Randomly? To see the spacing between impacts, you need to see at least two events on the screen at the same time if not more. To get a longer time span, you lower Fmax. And more often than not, you don't see the impacts in the time waveform. What happened?

The problem is the the antialiasing filter is filtering out everything about about 2x the Fmax frequency. When you set Fmax to a low frequency to extend the time span, all the energy resulting from the impact is filtered out!

Demod is one way around this problem. Then you can make the time record as long as you like and see a string of impact events. Feeding the accelerometer output into a strip chart recorder is another possibility.

Jon
Spintelligent Labs

PG,

You are not a dumby, you have just had a full chapter of intermediate vibration analysis thrust upon you in 5 or 6 formulas. It will make sense to you when you get the real training you want.

I didn't really check over your math, but I looks like you used cpm instead of Hz in the formulas. I didn't specify it in the first post, but did in the examples.

The data collection parameters that I suggested would be a good starting place, but continue to take readings with the setup you have now also. We need to know the sensitivity of the acceleromter you are using. It will be specified in millivolts/g and might affect your desired parameters.

Good Luck,

Danny

The sensitivity of the accelerometer is 100 mV/g. I think you asked what sample size I could get and it is 16384.
It appears that my program is not figuring right.
When I think 13 orders it thinks 238
When I think 27 orders it thinks 119
When I think 53 orders it thinks 59.

It looks like to me that by the pattern is more like the orders I figured. Shouldn't the pattern look more compressed when the orders are more and more spread out when the orders are less in number?

Thanks again

100 mv/g is a standard accelerometer. It will probably do ok for everything you are looking for here. The limitation will be in the natural frequency of the assembly of accel, cable and mounting device. If you are using a strong flat magnet on a smooth surface ou will get a linear signal out to about 150,000 cpm. You may need some glue-on targets and stud mounting for the higher frequency stuff. The maximum sample size is fine and it shouldn't limit you too much.

I'm not familiar with your software, so I can't help you with that other than to say that just about every time I find shaft speeds incorrect, I eventually find out what I did wrong in the set-up.

"Shouldn't the pattern look more compressed when the orders are more and more spread out when the orders are less in number?"

I don't understand your question, can you rephrase it?

You should notice some big differences with the added parameters. If you understand how, try collecting data with the parameters as off route data and post it.

Good Luck

Not that I didn't make a mistake but I double checked my set up and it appears to be okay. The attachment is of some data I took yesterday. If you look close you can see that the program does the opposite of what I think it should do as far as orders.

Thanks for you help.

Different_orders.doc (128 Kb, 34 downloads)

Getting back to your first question -- can you have too many LOR for a time waveform?

I always remember to rules of thumb.

1. A high Fmax and a low LOR = a detailed but short waveform and a spectrum with low resolution + gives a short data collection time.

2. A low Fmax and a high LOR = a high resolution spectrum with a long waveform with less detail + a long data collection time.

The circumstances dictate whether you need to lean towards #1 or #2. Go with #1 if you are taking a bunch of spectra during a quick startup/coastdown (maybe Fmax = 1000 Hz, LOR = 400). Go with #2 (classically Fmax = 200 Hz, LOR = 3200) if you are trying to discern between 2X and 2XLF on a two pole motor. Go somewhere in between if you are looking for impacts.

At any rate, the previous responses are more complete. Hopefully, remembering some general rules will help you.

Also, the main reason I wanted to respond was that I just received the November 2006 Uptime magazine. There is a nice article in there dealing with time waveforms. You can read the magazine at www.uptimemagazine.com.
Michael Titone

Mike - is that article available on-line? I went to the link and didn't see it (but maybe I didn't look hard enough)

Sorry. I guess you have to subscribe to be able to read the article. I only looked at the hard copy version when it showed up today. Then I checked a bookmark to get the website address. I didn't realize I had bookmarked some special area -- I guess only approved subscribers are allowed to read the article.

I definitely recommend subscribing to the magazine. It is one of the best free ones I have ever received + you can get a hard copy and/or an electronic version if you want.