How to condition tachometer (inductive pickup) signal w/ 2 signal components

[BradW]

We are acquiring a signal from an inductive pickup in a high speed turbine gearbox test article. The sensor is part of the system and we have to use this one. Current solution is to use an analog frequency to voltage converter (Daytronic 5D40) to generate an analog voltage read by the existing cRIO DAQ system.  Two problems exist:  There is significant whirl of the shaft, such that we are seeing the gear tooth signal superimposed on a lower frequency signal created by the motion of the gear as a whole, creating a 1/N frequency where N is the number of gear teeth,  Also, the gear frequency can change rapidly from 0 to 20000 Hz in as little as 500 ms, and we need to resolve the acceleration.  The analog F/V is very responsive and has special circuitry to detect and use 1/period algorithms when speed is changing fast, but it gets fooled when the whirl component (varies from unit to unit) approaches the same amplitude as the gear tooth component.  See the oscope shot:

The F/V will suddenly "lock" on the slower "beat" of the whirl and so the sensed and reported speed will suddenly drop to 1/N of its true speed, then jump back up, making the speed data noisy and difficult to use for test pass/fail analysis.

We would prefer a signal conditioning solution that would remove the whirl frequency without affecting the gear tooth frequency, but of course, the numbers change as the unit accelerates.  I've proven to myself that in the steady state case, we can easily detect the two tones in the frequency domain and reject the lower, but that doesn't allow lower speed response as it accelerates.  It spends too little time at any given frequency for an easy frequency analysis (too few samples for FFT with decenct df resolution). 

Does anyone have a pointer to existing signal conditioners that might help?

The pricey solution looks like a separate cRIO with analog acquisition of the time domain signal with custom algorithms to dynamically adapt to the slewing signals and extract the gear signal from the whirl signal.  This would be time consuming without good IP to do the up front signal analysis.  The existing DAQ system is running at 1 kHz and can't be reengineered to include the speed necessary for this algorithm (>40kHz) without very painful revision.

[Henrik_Volkers]

My first thougth was a differentiation of the signal, that should be able to lower the frequency point where the whirl amplitude is the same than the tooth amplitude......

A differentiation is a multiplication with 2*pi*f , since your tooth signal will be N times higher the frequency range of problems should decrease by 1/N. (or about it, since the amplitude is not only related to the frequency.... )

and you have a look at the noise....

feel free to post some critical data to play with ....  and some more timing needs....  for a production test you can sample the data and do the analysis later, for a turbine monitoring you what to prevent a run up 😄

[BradW]

You're right, post test analysis can sort out the issue, but as a production cell, there could be real workflow issues.  The safety concerns aren't really high as this is primarily a monitoring system.  Control and guard circuitry is separate.

I'm trying to avoid any significant software changes, which is why I posted here in signal conditioning.  If I could just use different front end conditioning and leave the rest of the system alone, I would avoid some painful work.  This is an aerospace production test facility, and the system and design is configuration controlled.  Changes, particularly software changes, have a long approval process both internally and with the customer.  We will be back in that cell doing some work soon so may have more complete data soon.

[Henrik_Volkers]

about timing needs and post processing for a test cell : if you apply Gaussien chirplet fits it migth take some ms (or say a half a second) for post processing and if I remember rigth the frequency resolution is not as strickly bound to the window length .....  have a look at the Sine curve fitting example in LV

An analog preprocessing would be a differentiation surrounded by two LP filters ..... but that will just shift the problem area to lower frequencies...

[johnsold]

Add another sensor which only picks up the the shaft whirl. Then you have a "clean" signal at the lower frequency. With appropriate amplitude and phase adjustments, subtract it from the combined signal before to gets to the "gullible" F/V converter.  Even without complete cancelation you may be able to reduce the whirl frequency component enough to keep the F/V happy. This approach does not require any tracking filters, differentiators, or other complicated circuits. 

Lynn

[BradW]

Unfortunately, there is no way to access the shaft other than with the tach gear in question.  These are production items, and we are obliged to maintain them in this configuration with no modifications.  This is heavily configuration-controlled qualification testing.  I may have to resort to the preprocessing like I thought I might, but I at least have some leads now from this thread on how I could do that in software/FPGA. 

There actually is an electronic speed control module which also uses this same signal to maintain control of the turbine, with it's own F/V circuit, but it gets around this glitchiness by filtering it pretty heavily. Unfortunately, I can't do that because my test has to have the dynamic range and resolution to define the system performance.  For a one-off test, that can be done by acquiring lots of time domain data and postprocessing, but is a pain for production testing.