Conclusion Get more connected.

A plot of Estimated value versus
Specified value is shown in Figure 3.
Prediction of OH number on the 8
independent samples using this
calibrated method gave OH numbers
with a standard deviation of 5.0.
Conclusion
The aim of setting up a NIR based
application is to provide an alternative
method to the wet chemistry reference
method with similar accuracy.
Analysis of OH number in polyols
is possible using NIR spectroscopy
and provides a considerably faster
route to OH number than the current
methods. From the calibrated methods
presented here, it can clearly be seen
that OH numbers covering a very wide
range can accurately be predicted to
ca. +/– 5 OH number. Over a smaller
OH range, this accuracy is increased
to ca. +/– 1 OH number which can be
as good as the reference method itself
therefore proving NIR spectroscopy
to be an extremely valuable, nondestructive
technique to the rapid
determination of physical and
chemical properties.
Factors Standard Error of Prediction % Variance
Wide Range 3 4.60 99.98
Figure 3: Estimated vs. Specified – wide range method.
Get more
connected.
www.perkinelmer.com
PerkinElmer Instruments
761 Main Avenue
Norwalk, CT 06859-0010 USA
Tel: 800-762-4000 or
(+1) 203-762-4000
Fax: (+1) 203-762-4228
U.K., Perkin Elmer Infrared
PerkinElmer and QUANT+ are trademarks of PerkinElmer, Inc.
Spectroscopy Business
Post Office Lane,
Beaconsfield, Bucks, HP9 1QA
Tel: (01494) 676161
Fax: (01494) 679331/3 © 2000 PerkinElmer, Inc. Printed in U.S.A.
D-5664A
KG010000

FT-NIR SPECTROSCOPY
application note
The Determination of OH Number in
Polyols Using FT-NIR Spectroscopy
Summary
NIR spectroscopy is an invaluable tool
for the quantitative analysis of a wide
range of chemical compounds. In
combination with a number of
chemometric methods, the technique
provides a fast, non-destructive route
to the determination of physical and
chemical properties.
This note describes the use of
FT-NIR spectroscopy in a typical
polyol analysis with an outline of
the development of the quantitative
method used for the application.
Introduction
Polyols are long-chain polymers
which contain alcohol functional
groups and are produced via
reactions involving organic oxides,
acids and multi-functional alcohols.
A wide range of products
including, surfactants, foams, paint
additives, and adhesives are
manufactured using polyols. The
production of polyurethanes, for
instance, involves polyol
intermediates.
Polyol products are typically
produced via batch reactor processes
held at high temperatures (>250˚C).
Current analysis of the polyols
produced normally take the form of
back titrations. The resultant OH
number is an average value over a
number of titrations.
This analysis generally takes
place in an off-site laboratory, often
far removed from the conditions
present at the production site. The
sample is extracted from the bulk,
and the production process halted
while the analysis is completed; a
process that can take several hours.
Care must be taken in handling the
chemicals involved, especially when
analyzing samples at elevated
temperatures. The possibility of
moisture being absorbed from the air
is highly detrimental to the analysis
of the OH value. A system whereby
the sample must be heated is applied
if the sample is a solid or highly
viscous liquid. Even when all these
factors have been minimized or
taken into account, the process still
depends on the analyst. The current
method for determining OH number
is expensive, time consuming, and
prone to human error.
A typical NIR based application
is faster, more precise and reliable
than other methods. It also reduces
the need to handle potentially
hazardous substances. The bands in
the NIR region (ca. 15000 – 3000
cm-1, 667 – 3333 nm) are primarily
overtones and combination bands
normally associated with C-H, N-H
and O-H bonds. Since organic
polymers are composed of carbon,
hydrogen, nitrogen and oxygen
atoms, the NIR spectra of polymers
feature sharp, strong absorbance
bands.
Polymer manufacturers can use
NIR spectroscopy to perform
analyses on the production site.
This has obvious advantages in the
area of quality assurance. NIR
analysis is also capable of process
control as the analysis of
intermediates occurs in real-time.
Since many polyol reactions are
irreversible, the analysis of such
intermediates by NIR spectroscopy
provides a commercially valuable
tool.
A typical NIR method
development for OH determination
consists of first assembling a set of
preanalyzed samples which must
span the OH range for which the
calibration is intended. The number
of samples used depends on the OH
range and accuracy required, but for
most applications in the range ca.
20 – 200 OH approximately 20
samples are adequate for calibration.

Once a set of reference samples is
obtained, it is advisable to divide these
into two sets; a set for calibration, and
a set to test the calibration often
referred to as the validation set. The
suitability of the samples used in the
calibration set is particularly important
since it must closely match the
product being manufactured at the
production site. A calibration set must
contain enough samples to enable the
estimation of the calibration constants,
and the range of the samples should
be representative of the range of
future samples.
A calibration, based on ‘ideal’
samples prepared in a completely
stable environment, may give very
small errors when used for similar
samples. However, when the
calibration is used to produce a value
for a real product, the accuracy is
often degraded. Furthermore, a tradeoff
exists between calibration sets
covering a very wide range of OH
number and the accuracy with which
OH number can be predicted.
Therefore, it has been found that the
accuracy increases when using a
method calibrated over a more
restricted (ca. 50) OH range.
The PLS (Partial Least Squares)
method available in the Perkin-Elmer
QUANT+ TM software uses the NIR
spectra recorded to derive a
calibration. This equation is then
used to predict the value of the
constituent of interest of future
samples using their NIR spectra. An
application that is analyzed by using
spectroscopy is more precise than
when measured by other methods.
However, the accuracy of the method
in predicting the true value of the
future samples is dependant upon
the accuracy of the reference method.
It is therefore important to stress
that the steps involved in setting up
the calibration are crucial to the
success of the NIR application. This
is particularly true for the collection
of suitable calibration standards,
and the analysis of the calibration
set by titration. This step should be
carried out with an objective of
producing the greatest possible
accuracy.
A suggestion for achieving this
accuracy is to perform the titration
in duplicate or triplicate. However,
it is more important to have as many
samples in the calibration set as
possible. In comparison, it is better
to collect twice as many calibration
samples than it is to perform the
reference analysis of the original
calibration standards in duplicate.
The most ideal, though costly and
time consuming, situation would
allow for the inclusion of a large set
of calibration standards, with each
sample analyzed in duplicate or
triplicate.
QUANT+ software was used to
derive a calibration matrix between
the calibration spectra and their
Figure 1: Extremes of calibration set.
reference OH values. The technique
of PLS extracts those spectral
contributions which are correlated
with OH value, and eliminates the
need to select specific wavelengths
for the calibration.
As previously stated, the near
infrared region consists of many
overlapping bands that contain
overtone and combination bands
involving the absorptions of mainly
OH, NH and CH bonds. The bands
of interest in the determination of
OH value are due to the OH first
overtone at 7150 – 6670 cm-1
(1399 – 1499 nm), the OH
combination bands of 5260 – 4760
cm-1 (1900 – 2100 nm), and
moisture content changes: ~5155
cm-1 (1940 nm).
Experimental
Fourteen standards of known OH
value were supplied to serve as a
calibration set. A further eight
samples were also supplied to act as
an independent validation set. These
were used to compare the reference
values with those values predicted by
the calibration equation. All samples
had been analyzed previously in
triplicate by the reference ‘wet’
chemistry method (titration).
The infrared spectra of all samples
were recorded on a PerkinElmer
FT-NIR Spectrometer. The samples
were scanned in an NIR quartz
transmission cell through a range
of 10000 – 4000 cm -1 at a resolution
of 8 cm -1 and strong apodization.
All spectra were recorded in ratio
mode using the shuttle accessory.
The shuttle automatically collects
alternate background and sample
scans to eliminate the effects of
water vapor in the air. A pathlength
of 5 mm was considered to provide
useful absorbance values. The
cuvettes were cleaned with acetone
and dried before refilling with the
next sample.
Two calibrated methods were
built using the reference data and
the spectra collected from the
FT-NIR Spectrometer. These were
setup using the QUANT+ software.
Both methods used the range 9000 –
4528 cm -1 with an upper threshold of
1.5 A. The offset option was
employed to minimize baseline drift

effects. The number of factors was
set between 2 and 6. The validation
option was set to Full Cross
Validation (see later), and the Expert
Assist (see later) was selected for the
calibration. The first method was built
using the full range (20 – 270 OH
number) of samples. However, the
second method was built covering a
much smaller range of OH number
(25 – 60). The second method was
able to predict more accurately
values within its low OH range. The
first method however, was able to
predict values better for the samples
lying higher in the range, but was not
as accurate for the mid-range values.
Results and discussion
The absorbance spectra for the
polyol standards and the unknowns
were collected and two examples,
representing extremes of OH number,
are shown in Figure 1. The spectral
variations at approximately 7200 –
6670 and 5300 – 4760 cm -1 (1389 –
1499 and 1887 – 2100 nm) clearly
show differing OH number, and
correspond to polymeric O-H first
overtone and O-H combination bands
respectively. Temperature variation is
important to consider in the analysis
of polyols and other NIR applications.
This application did not require
temperature control, and heating
was not necessary as all the samples
were liquid at room temperature.
However, care was taken to ensure
that the sample temperatures were
allowed to equilibrate and the
samples were left in the beam for the
same length of time before each
measurement was made.
The QUANT+ software is capable
of modeling the concentrations in
terms of the spectral contributions
relevant to OH determination. This
algorithm can extract the factors in
the calibration that are not responsible
for the specific OH value (the
chemical constituent of interest)
by eliminating the corresponding
dimensions (spectral absorptions)
from the model.
The error present in an OH
evaluation due to hydroxyl band of
moisture absorbed in the sample can
then therefore be minimized by
Factors Standard Error of Prediction % Variance
Limited Range 2 0.91 99.86
Figure 2: Estimated vs. Specified – limited range method.
digitally separating the different OH
bands using this feature of the
software. The calibration equation is
modeled to exclude irrelevant bands
from the analysis, and minimize the
error in the prediction results.
The Full Cross Validation feature
used in this application further
improves the reliability of results.
The software drops one sample from
the calibration set and performs an
entire calibration with the remaining
samples, then makes a prediction for
the sample left. In normal
applications the sample being
predicted is not present in the
calibration set, this gives a more
realistic if less accurate result. The
Expert Assist option in the software
makes intelligent decisions
regarding the spectra in the
calibration set. The Expert will
decide whether to reject any spectra
or specific regions of spectra in the
calibration set, or disregard any
property values on the basis of a set
of rules developed for IR quantitative
analysis. It will also identify any
outliers in the set. Once these
decisions have been made to exclude
some data in the set, the software
will decide whether or not the
calibration needs to be repeated.
The procedure continues iteratively
until all the rules are satisfied.
The method built over the narrow
range of OH number was used to
predict the samples of OH value
ranging from 20 to 50. The error
estimates are considerably better
than the results obtained using the
full calibration set. By looking at the
results tabulated in the table in Figure
2, 99.86 % of the data can be
explained using 2 factors with a
standard estimate of prediction (SEP)
of 0.91 OH number. Figure 2 illustrates
the accuracy of the method in the
form of an Estimated value versus
Specified value plot. Prediction of
OH number on the 8 independent
samples using this calibrated method
gave OH numbers that were found to
lie within ca. +/– 1 OH number of
the value obtained from the
reference method (titration).
The full range method uses a larger
calibration set to predict OH number
over a much wider range (20 – 275).
The values recorded in the table in
Figure 3 show that in this method,
99.98% of the data can be explained
using three factors in this analysis.
The SEP is much larger in this
method (4.60 OH number) and was
found to be better at predicting
OH number of the higher valued
unknowns.