Filter
Top Products
Figure 2
Bivariate Fit of Log Days by Inverse Temp
—Linear Fit
Linear Fit
Figure 2 shows a straight-line extrapolation to 200
years. However, because the paper is more stable
than that in Figure 1, only three data points are
available. Because of noise, the three points are
scattered around the predicted straight line. With
only three points and the higher noise, the
correlation is lower and the error band is higher. So
while the straight-line extrapolation predicts a
performance of 200 years, the statistically
calculated error is very wide and quite nonlinear.
According to the plot, the paper could last as long
as 600 y
ears or as few as 50. In this example,
running the test for a longer period of time would
e
v
entually g
enerate a fourth data point to give a
more precise prediction with less error around the
pr
ediction.
In light-fade testing, using only highly accelerated
tests can lead to significant errors because of
r
ecipr
ocity and other external factors. For example,
we now know that some inkjet materials are highly
susceptible to image fading due to very low levels of
ambient ozone in the atmosphere.
1
Consider the
outcome when only high-intensity light-fade
testing, which predicts a material’s stability over
1
00 years, is used. Because the test is so highly
accelerated and run in such a short time, it cannot
detect the impact of ambient ozone. Only after
images were displayed in the real environment,
which included low levels of ambient ozone, was the
t
rue stability determined. Then it turned out that
product stability was measured in weeks, not years.
As Kodak added new resin stabilizers to increase
the stability of the paper base, Kodak scientists
encountered another example of a test becoming
difficult to use because of noise in the data. The
then-current accelerated test to measure cracking
and crazing of the resin was no longer generating
the normal response of cracks that would predict
long-term resin stability. Instead of trying to
accelerate the test further and risk the creation of
even more noisy, possibly erroneous data, Kodak
scientists looked to the molecular level for a
predictive mechanism. They made measurements of
the actual decay of the resin molecules and found
that those measurements provided a good
correlation to the prediction of long-term resin
stability.
In these times when companies, including Kodak,
introduce new products at a faster pace, it is
tempting to run image-stability tests for shorter
times with greater acceleration and noisier data. It
would be even worse to use the high end of the
statistical noise to promote a product as more
stable than it really may be.
A
t Kodak we do neither. We continue to run highly
accelerated light- and thermal-fade screening tests,
using high light levels and high temperatures.
How
e
v
er
, we verify the results with less accelerated
tests by using lower light levels and lower
temperatures. Although these tests take
significantly longer, often more than a year, the data
is much more reliable. We use only this data to
support product claims for image stability. We take
the e
xtra time to generate reliable data because
3
log days
Inverse Temp
.
0034.0033.0032
.0031
.
003.0029.0028.0027
5
4
3
2
1
24°C
y
ears
200
1
D.E. Bugner, D. K
opperl, and P
. Artz, “Further Studies on the Apparent
Reciprocity Failure Resulting from the Accelerated Fade of Inkjet Photographic
Prints,”
Proceedings of IS&T’s 12th International Symposium on Photofinishing
Technology,
2002, pp. 54–57.