Pressure Transducer Signal Jumping

[kinghuds]

They are strain gauge based, this is the info I found for mV/V:

mV/V Specifications

Output:

mV/V, 100 mV @ 10 Vdc

(Ratiometric 5 to 10 Vdc)

Supply Voltage:

10 Vdc

(5 mA @ 10 Vdc)

Input/Output Resistance:

5000 Ω

±20% typical

The two I have both output from -5V to 5V, which I have calibration data for and is scaled in MAX.

[johnsold]

So, now we can make some deductions.

1. With a bandwidth of 1 kHz (and that is probably the -3 dB bandwidth) you will need a sampling rate > 2 kHz to meet the Nyquist criterion.  Given the likely slow roll off of the frequency response, the 10 kS/s recommended by Henrik Volkers should be considered a minimum.

2. If you are only interested in slowly varying phenomena, a low pass filter could reduce the sampling requirement.

3. With an excitation of 10 V the sensors probably have a common mode output around 5 V and a total signal range from 4.9 to 5.1 V (Common mode +/- differential mV/V).  Suitable differential signal conditioning would eliminate the common mode voltage, amplify the signal voltage, and possibly provide bandwidth limiting and anti-aliasing filters.

4. The PXIe-6341 has 16 bit sampling, but your effective resolution on the signal is ~9-10 bits if the numbers in item 3 are close and you are not using any signal conditioning.

What are you doing with the data after you make the measurements? What is the purpose of the whole system? This will help us advise you.

Lynn

[kinghuds]

Right now the system is just using these two sensors to measure pressure differentials through different ducts, orifices, etc. So the needed rate would only be 5-10 points a second just to give improved measurements over the manometer we are currently testing with and to allow for the data to be recorded from testing as well. From what I'm reading my approach to getting this rate by actually setting the sampling rate that low is really off.

I'm also doing the same thing with some of the load cells I am working with, wich give me 0-30mv at 10V excitation.

[johnsold]

I woudl buy or build a signal conditioner which removes the common mode voltage, filters the signal at 5 Hz, amplifies it to match the range on your DAQ device and then sample at about 20 samples per second.  Average 2-4 adjacent samples and use the averaged values.  I expect that this would closely match your manometer.

Lynn

[Ben]

NI sells strain guage modules that hve worked very nicely for me. They have built-in nulling and the bridge completion circuitry.

Ben

[Broken_Arrow]

If the software filter is available for your card, it may be worth a try. On the torque (strain) sensors I use, it helps a lot.

[kinghuds]

With the load cells I'm using the TB4300 blocks and they actually are reading consistently, I just can't get constant readings with the pressure sensors. Is a hardware filter the only way? Or is there a way to get a less frequent averaged reading wholes using a high rate like 10ks?

[johnsold]

If you sample fast enough to satisfy the Nyquist criterion on the noise (which is everything about about 5 Hz out to > 1 kHz), then you can filter in software.

If you sample at 10 kS/s and average blocks of 1000 samples you would have 10 points per second.  Unless your duct is whistling at several kHz, you probably do not have much noise energy at higher frequencies.  It would not hurt to look at the raw 10 kS/s signal on a graph or to do an FFT on it to see if it contains any high frequencies.

Lynn

[kinghuds]

What is the best way to get the code averaging every x samples? Or would something like a running average of the last thousand samples be smoother as well? I don't have LabView at home so I'm not exactly sure how to approach it.

[johnsold]

If you set your data acquisition to read 1000 samples at a time (with the rate set to 10000 samples/second), all you need to do is to average that block of 1000 samples as soon as it is read.  You can use the Mean.vi in the Mathematics >> Probability & Statistics palette or use Array Sum followed by Divide by 1000.

A running avaerage is a bit more complicated to program but would probably produce a smoother result.  In intermediate approach might be to read 500 samples at a time and keep two blocks of samples so that you average 1000 samples every 50 ms, producing some smoothing due to the overlap.

Which is the most appropriate may depend on how fast your data changes.  How much change do you see on the manonmeter from one reading to the next? Reading 10000 samples and averaging them (once per second) should produce results similra to the manometer.

Lynn