Precision and uncertainty of video-extensometer

[JoacimR]

Hi all

This is my first post after a few years absence from Eng-Tips. Hopefully I'm posting in the correct forum...

I'm new to extensometer measurements so I wanted to find out more about the measurement uncertainty of my video-extensometer.
Performing the uncertainty calculations described in a paper available from the optics manufacturer of the extensometer system, I get an measurement uncertainty in the order of 1 millimeter for measuring the position of an object.
The supplier of the extensometer system states the precision to be in the order of 10 micrometer for the strain measurement.

How this is possible?
Am I way off believing that the uncertainty needs to be at least one tenth of the measurement precision?
Maybe there's another paper "out there" explaining this?
Is it all due to the extensometer software performing some image processing techniques which reduces the measurement uncertainty?

Any input on this is most welcome.
Regards,
Joacim

 

[btrueblood]

Possibly it's because strain is a ratio of lengths, and the scale errors cancel. I.e. things like parallax error or range error will result in a constant offset of the measurement of L0 and L1, and this offset will cancel out when you calculate strain = (L1-L0)/L1. Just a thought, you would have to verify that the biggest part of your 1mm error is from such constant effects, and that the repeatability of measurements in a single setup is as good as 10 micrometers.

As far as uncertainty, no - you can have the precision of the instrument be much greater than its uncertainty. One might round the readings, or better, report the uncertainty, when reporting/analysing results. When calibrating the instrument, however, you should have the calibration standard have an uncertainty better than the instrument's uncertainty, and generally better than 1/10.

 

[IRstuff]

The classical desired test accuracy ratio (TAR) is 10:1, ala MIL-HDBK-52B, but cutting edge instruments tend to be pushing the calibrations standards, so often, one can only achieve TARs of 4:1 or even less.

What size is the object and what is the resolution of the camera?

TTFN
faq731-376

 

[IRstuff]

Or more germanely, what is the projected object plane dimension of a pixel?

TTFN
faq731-376

 

[JoacimR]

Thank you for your quick replies!

btrueblood,
I agree with you on this. But how do I compensate for the error given by the angle from the line of sight?
See the attached picture.
As the material is stretched the beta angle increases and the end of L2 approaches the boundary of the FOV. Then the measured distance, y, is smaller than the actual distance, L2.
Using L=0.5 m, L1=10 mm and L2=150 mm I get an error, L2-y, of approx. 7 mm. If I'm calculating this correct, it's a very large error.
Is this kind of error usually compensated for by the extensometer software?

IRstuff,
At a distance (L) of 0.5 m my resolution is 160 micrometers/pixel.

 

[btrueblood]

Joacim,

Your picture doesn't make a lot of sense for operation as an extensometer, i.e. making a strain measurement in a tensile test. I.e. (based on your figure) you are measuring L1 and then L1+L2 in a tensile test, right? And typically, L2<<L1, so right off the bat, the parallax error could be ignored for small strains. To be more specific, the error in measurement for L2+L1 and L1 could be expressed as a constant factor, like:

L1 = L1measured * f1
L1+L2 = (L1 + L2)measured * f2

and for small enough displacements, f1 ~ f2

Thus when calculating strain, the "f's" would cancel out.

But let's say your figure is a close approximation of the measurement you want to make. There are corrections that can be applied to the measurements, if you know (or can approximate) the angles involved.

Second, if you know where the displacement is going to happen, you can adjust the camera field of view to minimize the parallax error for that area.

 

[IRstuff]

> I don't think that your analysis is correct. Assuming they halfway know what they're doing, they would ostensibly have a correction, either because their optics are fully telecentric, in which case the error should be comparable to the ability to resolve the actual edge of your object.
> Even if it were true, you could certainly halve the error by placing the the LOS at the virtual center of your setup. Additionally, you could certainly calibrate out the error by using a reference scale like a steel ruler.
> Finally, if the measurement system is ostensibly insensitive to any distance within its operating specifications, then it is sufficiently telecentric to do the measurement accurately.

TTFN
faq731-376

 

[btrueblood]

I should have said, I have no idea if any of the corrections are built-in to your extensometer or not, and no idea if a telecentric lens automatically compensates for parallax error or not. I glanced briefly at the link the OP sent, and then got bored and ignored it.

IRstuff's suggestion that you calibrate to a known standard is an excellent one.

 

[JoacimR]

Thank you for your input on this. I will compare the extensometer to a LVDT to make sure I get this right.

 

[Compositepro]

An LVDT is another extensometer which itself must also be calibrated with an extensometer calibrator. The calibrator is a super accurate dial micrometer.

http://www.instron.us/wa/acc_catalo...r Extensometers&ref=http://www.google.com/url

 

[btrueblood]

Or a set of gage blocks?