PGA970: LVDT sensors and PGA970

Hello Jean-Francois,

Can you share some more information about your application? What are your linearity requirements and what kind of accuracy have you been able to produce so far? Are you having trouble with linearity across position at a single temperature, or is the problem across temperature?

Can you share your schematic? It would be helpful to see the LVDT sensor you are using and the surrounding circuitry. If you would prefer not to share publicly you can add me as a friend and share it as a private message. 

We don't have an algorithm example available, but we can review the analog portion and the PGA970 device setup to make sure there isn't anything there affecting the accuracy.

Regards,

Hello Scott,

Some details on our project:

We are working on a metrology application with a high quality 5-wires LVDT sensor. The displacement range is 10 mm.

The schematic we are working on is the schematic of your EVM. The PGA970 delivers a pure sinusoidal signal, frequency 5 kHz, amplitude 5 V after amplification.

We note on the secondaries windings of the LVDT an unbalanced signal (relative to ground), probably caused by a capacitive unbalance of the windings. We suppose the two differential amplifiers of the PGA970 can deal with this unbalance.

All seems fine.

With the ratiometric method S = (A-B)/(A+B) we obtain easily  a 1% accuracy on the results  (ie 100 µm), but we need a much better accuracy (target: better than 5 µm).

Our first attempt to linearize the signal is to use a polynomial correction. We get a 20 to 30 µm accuracy. We cannot obtain a better result due to a slight asymetry in the curve.

Presently, we are working on a multiple segment method (21 segments) with slopes defined independently for each segment. This process seems very fast.

Can you tell us if we are doing something wrong ?  Do you know a better method to get accurate measurement ?

We do not use presently the phase information. Do you think the phase information can help in the calculation algorithm ?

Thank you for your answer.

Hi Jean-Francois,

I think your next step of trying an algorithm with sloped segments is the best choice to address the accuracy issue. I do not have any recommendation beyond that as far as the linearity compensation algorithm is concerned.

As long as the PGA970 is operating within the recommended operating conditions, then it does not seem like you are doing anything wrong. It may just be that this particular LVDT has some linearity issues that are difficult to account for.

Unless you need to know when the LVDT core has passed the central position you should not need the phase information.

Regards,