order analysis toolkit. Problem converting to even angle signal

[QiaoGuo]

Hi Crimolvic

Let me supplement the delay of OAT Analog Tacho Process.vi and OAT Convert to Even Angle Signal.vi.

In order to get a non-empty even-angle signals output, the minimum number of needed input pulses (only equals revolutions when pulse/revolution=1) is summarized for each case as below

A. If 2 .5*max order<pulse/revolution, the minimum number of input pulses needed is 4
B. Else, Fast mode, zero delay = True, the number is 5
C. Else, Fast mode, zero delay = False,  the number is 7
D. Else, Accurate mode, zero delay = True, the number is 28
E. Else, Accurate mode, zero delay = False, the number is 53

And about what you did, "insert false pulses" to avoid a longer acquisition, I am sorry that I do not quite understand that.

Would you please make it clear so that we can help you?

Thanks.

[crimolvic]

Hello QiaoGuo,

thanks for the info. Once again I think it should be included on the manual.

Let's say I want to get an order spectrum with resolution 1/20 Order. Let's also say that whe use the condition E (Accurate mode, zero delay = False). If I don't want averages, then I'd need to acquiere (20*(pulse/revolution)+53) pulses. Let's say pulses/revolution=1, then I need to acquiere the signals with a length equivalent to 73 revolutions. However only 20 of them will be used for the spectrum (app. 73% of the measurement is lost).

I though on aquiring the signals with a length equivalent only to the 20 revolutions needed (shorter acquisition time) and then programmatically "insert artificial pulses" (maybe false was not the best word) before and after the 20-revolutions signal. I did that inserting 26 pulses before and 27 pulses after, but the resampled waveform obtained was not satisfactory (I thought because of the causality/non-causality characteristics of the filter). The artificial pulses are made so that the (artificial) RPM is constant.

Now that I see the condition D of your post, it might be that I should insert 25 and 28 before and after respectively, and not 26 and 27. I'll try that next friday.

Have you ever try this?

[QiaoGuo]

Hi Crimolvic

Well, for order-domain calculations, it is common to lose several portions of samples before and after the output ones.

But the missing samples also count in the final output results, which means the output results are dependent on all the input samples, including both the missing ones and the output ones.

In your mentioned case, 73% is calculated out when your input is 73 revolutions and with the expectation of a 0.1 resolution's Order Spectrum. For an on-line MCM applications, the percentage of data loss is very very low.

In a constant speed case, inserting artificial pulses sounds reasonable.

But you have to meet the condition that, between your inserting artificial vibration signals and the actual acquisition signals, every order component signal has to be continuous in phase, otherwise, although the output revolution indexes are what you expect, the results are totally incorrect. Furthermore, it is impossible to assure the phase continuity when inserting artificial vibration signals for the randomicity. 

In conclusion, for the existing of several filters inside our VIs, I am afraid to admit that a little bit long acquisition time is inevitable.

The only way to avoid this is to correctly set the "analysis method" and "Zero Delay?" according to your specific use cases.

Hope my answers can help you.

Message Edited by QiaoGuo on 02-27-2008 11:25 PM

[crimolvic]

Hi QiaoGuo,

thanks for your answers, of course they help!

please 3 more questions:

1. By setting "Zero Delay" = True the same filters are used, but the filter delay is arbitrarely setted to cero (0). Does this means that the "corrupted" data at the beginning is also used for the calculations of the spectrum?. If yes, I wouldn't use the "Zero Delay" = True in no case. I understand that this filter delay is necessary because of the filters that you use.

2. Now regarding the pulses at the end: On a one-shot application these are not used. Then a shorter acquisition time could be acheieved by modifying the vi's in order to not to buffer the last portion of the data to be used by an unexistant new measurement. Unfurtinately I can not say if this is possible to do or not, bacause I think this process is performed through a dll.

3. When one use the accurate mode, then two filters are used for the interpolation, being the second a CIC filter. Is the first filter a FIR? Why does this previous filtering improve the accuracy?

Thank you for your time,
crimolvic

[QiaoGuo]

Hi Crimolvic

These are my answers for your three questions.

1. The answer is yes, that's why we make the default value of "Zero Delay?" to FALSE. And the filter delay is not "arbitrarily" set to zero, the delay between FALSE and TRUE is also calculated by the filters inside.

2. Just the same in my last post.

"the missing samples also count in the final output results, which means the output results are dependent on all the input samples, including both the missing ones and the output ones."

The missing pulses at the end and at the start both determine the final output results. I think it is impossible not to buffer the last portion of the data for shorter acquisition time.

3. As it is known, the CIC filter is an efficient way of perfoming large rate decimation and interpolation. But there is one drawback that its passband characteristic is not so perfect. Thus in accurate mode, before the CIC interpolation filter, we uses three interpolation filters, each of whose interpolation factor is 2, in order to improve the passband characteristic for a more accurate result. That brings the extra delay than the fast mode.

If you are interested in this, you can use Digital Filter Design toolkit or something else to see the frequency response differences between the CIC filter in fast mode and the cascaded filters in accurate mode.

[crimolvic]

Thanks QiaoGuo for all your support!

(and sorry for bothering you with so much questions...)


Message Edited by crimolvic on 02-28-2008 07:41 AM