i1Pro 2: Amount of deviation between measurements

[MichaelKoerner]

Please help me to shade some light on the following phenomenon, if possible:

Since about a year, I use an i1Pro2 to measure printed test charts. Recently I started to analyse/compare multiple measurements of the same chart. Using Babelcolor Patchtool, I find deviations at some patches larger than 1 dE*, some even larger than 5 dE*.

Looking at the position of those deviating patches at the test charts, I find no reproducable pattern.
Looking at the color/lightness of the spikes, of course many of them are rather dark, but I also see strong deviations at light blue or orange patches, for instance.

I measure my test charts with i1Pro2 in i1Profiler, manually, using the measurement ruler. Of course I try to keep variations during measurements as low as possible.

At an average test chart containing 1600 patches I find 1-10 patches with deviations larger than 1 dE* and 4-50 patches with deviations between 0,5 and 1 dE*, generally higher devations on glossy media.

Here are my questions:

• Are spikes like this "normal"?
• If not, are those a sign for technical malfunction or user error?
• Concerning user error: Where can I find USEFUL best practices concerning repeatable measuring techniques with i1Pro (besides X-Rites youtube channel)?
• Concerning technical malfunction: Are there alternatives to X-Rite to send the device to - as X-Rite charges ridiculous prices... My device is just 13 months old, so young, but unfortunately out of warranty :-(

Please find some screenshots from Patchtool's comparison function below. Patches are sorted top-down/left-right, do the patch with the highest deviations sits top left, the one with the lowest bottom right.

Thanks in advance for any ideas/input!

[DP]

https://forum.luminous-landscape.com/index.php?topic=108238.0

"...But if the bulb gets warmer still, say by being used for strip
readings, then the tungsten gets re-deposited on the filament,
and the bulb gets "cleaned". So after some extended strip readings
(i.e. several measurements in a row, made with the button down
and the lamp on for as long as possible), the instrument no longer
displays such dramatic error increases with temperature. In fact
after "cleaning", the worst I could get was 0.08 Delta E, a 20x reduction,
and back in line with instrument specifications...."

[JRSmit]

What patch size did/ do you use to print the patch charts?

[MichaelKoerner]

First, sorry for the late reply.

I infer from your answers that the deviations I described are not "normal". That was my main question...

Now, before sending my almost new device to x-Rite (and paying a fortune for checking), please help me finding a possible reason:

@DP: Thanks for pointing me to this very interesting thread! Unfortunately, I don't do spot measurements, strips only - so I don't think Graeme's observation describes the phenomenon happening at my place. Besides, patches showing deviations higher than 0.5 dE* are distributed over the whole target without a (at least for me) recognizable pattern. I attached a screenshot representing the actual layout of one target for clarification, deviations >0.5* marked yellow, >1* marked red.

@Jan: I am aware of the need to use patches of min. 8.6 mm size to fulfill ISO standards. Since I recognized that phenomenon (some weeks ago) I even used 9mm (height and width) and tried to scan as slow as possible (taking about 4 sec per line). Same results - less deviations, but still some.

For a quick test, I compared some of the patches showing very high deviations visually. To be honest, I could not tell a difference, so perhaps I will decide just to live with it... Nevertheless: Any more ideas very welcome!

Have a nice weekend, Michael

[DP]

@DP: Thanks for pointing me to this very interesting thread! Unfortunately, I don't do spot measurements, strips only - so I don't think Graeme's observation describes the phenomenon happening at my place. Besides, patches showing deviations higher than 0.5 dE* are distributed over the whole target without a (at least for me) recognizable pattern. I attached a screenshot representing the actual layout of one target for clarification, deviations >0.5* marked yellow, >1* marked red.

how about you do a number (like 20) sequential measurements w/o much interruption measurements of the target (make it of course way less than 1600 patches for simplicity), then discard 10% top/bottom outliers for each patch and present what will be deviations afterwards ... while spot reading is naturally affected more - it is still possible to get this w/ strip reading - techniques vary ...

[DP]

I attached a screenshot

where is it ?

[DP]

PS: another idea (because why not) - may be your specific device does not feed the bulb with sufficiently stabilized current/voltage - hence random (if they are truly random with each new reading of the same target) errors ?

[MichaelKoerner]

where is it ?

Ah, sorry - did not work the first time. Now it should be visible.

how about you do a number (like 20) sequential measurements w/o much interruption measurements of the target (make it of course way less than 1600 patches for simplicity) ...

Yes, I will do tests in that way - thinking about using a TC2.83 target (294 patches) and 10x10mm patches.

then discard 10% top/bottom outliers for each patch and present what will be deviations afterwards ...

Hm... Not sure whether I understand exactly what you mean by that. Those outliers are the exact thing I want to analyze - discarding them seems to be counterintuitive to me?

[Rhossydd]

When I first got an i1Pro2 I also ran some repeatability testing. I’d say that your results are broadly similar to what I see here.
Does i1Profiler’s comparison function report the same errors as Patchtool ?

My figures from profiler are below, I’ve added the data from my Isis for comparison;
Average dE
all patches    0.46      (isis 0.1)
Lowest 90%   0.39       (isis 0.07)
Highest 10% 1.23       (isis 0.29)

Max dE
All       3.4       (isis 1.29)
Lowest 90%    0.86       (isis 0.18)

The really important numbers are the overall averages, rather than the few outliers. I’m sure I’ve read in the past that there is a degree of error checking within profiler when building a profile and random outliers get ignored to ensure overall smoothness of the profile.

4s per scan sounds rather slow. Watch the X-Rite tutorial videos and try to copy their scanning speed. I’d suggest twice as fast at least and that might improve your success rate.

If the profiles you've built look OK, I wouldn't worry about getting too obsessed by errors that are ignored by the software anyway.

[MichaelKoerner]

Thanks for your very clarifying post!

[DP]

Hm... Not sure whether I understand exactly what you mean by that. Those outliers are the exact thing I want to analyze - discarding them seems to be counterintuitive to me?

when I do measurements I always do multiples and discard outliers and then average the rest of the results and use that... never just one reading per patch... to use just one measurement for a patch or to include outliers shall be counter-intuitive !

[MichaelKoerner]

DP, I understand now - and this info accords with Rhossyd's post.

In my view, it's a shame such vital information does not come along with x-rite's (hardly existing) documentation. So glad this forum and it's experienced users exist! Thanks once again.

[Rhossydd]

never just one reading per patch

Just remember that the i1 makes multiple readings itself.

[MichaelKoerner]

4s per scan sounds rather slow. Watch the X-Rite tutorial videos and try to copy their scanning speed. I’d suggest twice as fast at least and that might improve your success rate.

Just remember that the i1 makes multiple readings itself.

Until I got aware of the deviations I used the scanning speed you mentioned - about 2 sec per row, as seen in the X-Rite videos. Searching for a reason for the outliers I remembered i1Pro's sample rate of 200 Hz and reduced my speed down to about 4 sec per row. Thus I hoped to give the instrument more time to gather more measurements per patch. In my opinion the slower pace helped getting more consistent results, but this statement would need more tests for verification (BTW: One can "feel" the resulting increase of calculating time as it takes longer until the green light on the i1Pro2 appears, telling you can switch to the next line).

I suppose this is the kind of multiple readings you refer to?

[Rhossydd]

I suppose this is the kind of multiple readings you refer to?

Yes.
I remember initially thinking that slower ought to be better, but it doesn't seem to work that way in practice. I'd assume that there is a specific sweet spot, but from what I can remember of my tests anything around two to four seconds didn't make much difference to repeatability.
I also recall that if you take far too long or flash it over the strips you'll get an error message anyway.

[MichaelKoerner]

X-Rite suggests to perform multiple measurements (in several posts on it's website) and to let i1Profiler average the results. So I assume that it makes more sense investing one's time into more measurements at a higher speed than to do less measurements with low speed.

BTW, i1Pro 2's Zebra strips should compensate for speed changes and/or interruptions...

[Doug Gray]

BTW, i1Pro 2's Zebra strips should compensate for speed changes and/or interruptions...

The Zebra stripes are used with dual sensors and provide quadrature position sensing which the i1Pro2 uses. This allows relative position sensing to < .5mm an eliminates things like jerky motion or brief interruptions in speed. It's quite effective. So much so that you can even slide the puck then reverse it a patch or two and continue forward to the end of a strip and i1Profiler will properly read the patches. The spectro samples are aligned with the quadrature readings which are scaled to the two thin black stripes at the end of each row.  Much better than the old i1Pro.