Final Report
submitted to
NATIONAL AERONAUTICS AND SPACE ADMINISTRATION
GEORGE C. MARSHALL SPACE FLIGHT CENTER, ALABAMA 35812
November 1, 1992
for Contract NAS8 - 38609
000
Pr_
Z
uJ_Jq.p _.
0
-_acct)
.J ¢:
0.._ LL
UJ
q:_ o')
•..J c_
@
r-
ey 2' _
U¢
N
WD
rn
0
N
Delivery Order 18
entitled
Materials Surface Contamination Analysis
by
Gary L. Workman Ph.D.
Principal Investigator
and
William F. Arendale, Ph.D.
Consultant
In-line Process Control LaboratoryCenter for Automation & Robotics
University of Alabama in HuntsvilleHuntsville, Alabama 35899
https://ntrs.nasa.gov/search.jsp?R=19930007811 2020-08-01T17:53:23+00:00Z
Table of Contents
Page
1.0
2.0
3.0
4.0
5.0
6.0
6.1
6.1.1
6.1.2
6.2
6.3
6.4
6.5
7.0
8.0
9.0
APPENDIX
Introduction ............................................................. 1
Research Objectives ...................................................... 2
Optical Fiber Spectrometry Concepts ..................................... 3
Brief Description of Chemometric Principles .............................. 9
Experimental Approaches ................................................ 1212Results ...................................................................
Spectral Determinations of Contaminants ................................. 13Matrix of Contaminants .................................................. 13
Choosing the Reference Spectra .......................................... 19Chemometric Determination ............................................. 20
Discussions Related to Two Models ...................................... 23
Observations Related to I-ID2 on D6AC Panels ........................... 27
Observations related to the reflected energy from the surfaces ............. 55
Conclusions .............................................................. 58
Optimization of the use of NlR for quality assurance ...................... 58
Acknowledgments ........................................................ 5961
.................................................... . ...............
Table of Figures
Page
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
°
2.
3.
4.
5.
6.
7a.
7b.
7c.
7d.
7e.
7£
7g.
7h.
7i.
8.
9.
lOa.
10b.
11.
12a.
12b.
Spectral Domain of Interest .........................................
Water and carbon dioxide absorption bands .........................
Near-infi'ared spectra of water and deuterium oxide. 0Neyer) ........
Spectrum of Water in lmm Cell (UAH) .............................
Spectrum of acetone in lmm cell (UAH) ...........................
Example of Reflection/Absorption Process ..........................
UV Spectra of Contaminants ........................................
PCA Analysis of Y matrix ..........................................
PLSR of Y and X (UV data) ........................................
Regression Analysis for HD2 .......................................
Regression Analysis for WD40 ......................................
Regression Analysis for Lubriseal ...................................
Regression Analysis for Apiezon ....................................
Regression Analysis for TapMagic ..................................
Regression Analysis for Sebum .....................................
Spectrum 65 Smooth vs Noisy ......................................
Coupon E8A 7 Days at 100°F/60% RH (Front Side) ................
Observed vs Predicted times for samples in HIT series ...............
Observed OSEE values vs predicted values in HIT series ............
Using Laboratory Data Model SO C to predict HIT and E4 sets .....
Model NAR Loadings Plot for Factor 1 .............................
Model NAR Scores Plot for Factor 1 ................................
12c. Model NAR Loadings Plot for Factor 2 .............................
12d. Model NAR Scores Plot for Factor 2 ...............................
12e. Model NAR Loadings Plot for Factor 3 .............................
12f. Model NAR Scores Plot for Factor 3 ...............................
12g. Model NAR Loadings Plot for Factor 4 .............................
12h. Model NAR Scores Plot for Factor 4 ................................
12i. Model NAR Loadings Plot for Factor 5 .............................
12j. Model NAR Scores Plot for Factor 5 ...............................
13a. Loadings Plot for Factor I ..........................................
13b. Scores Plot for Factor 1 ............................................
13c. Predicted with 1 Factor .............................................
13d. Loadings Plot for Factor 2 ..........................................
13e. Scores Plot for Factor 2 .............................................
13f. Predicted with 2 Factors ............................................
13g. Loadings Plot for Factor 3 ..........................................13h. Scores Plot for Factor 3 ............................................
13i. Predicted with 3 Factors ............................................
13j. Loadings Plot for Factor 4 ...........................................13k. Scores Plot for Factor 4 ............................................
3
5
6
7
7
8
16
16
16
17
17
17
18
18
18
22
24
25
25
26
32
32
33
33
34
34
35
35
36
36
37
37
37
38
38
38
39
39
39
40
40
ii
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
131. Predicted with 4 Factor ............................................
13m. Loadings Plot for Factor 5 .........................................
13n.
13o.
13p.
13q.
13r.
13s.
13t.
13u.
13v.
13w.
13x.
14.
15.
16a.
16b.
17a.
17b.
17¢.
17d.
17e.
17£
17g.17h.
17i.
17j.
18a.
18b.
18c.
18d.
18e.
18£
19a.
19b.
19c.
19d.
19e.
19f.
19g.19h.
19i.
19j.19k.
191.
Scores Plot for Factor 5 ............................................
Predicted with 5 Factors ...........................................
Loadings Plot for Factor 6 .........................................
Scores Plot for Factor 6 ............................................
Predicted with 6 Factors ...........................................
Loadings Plot for Factor 7 .........................................
Scores Plot for Factor 7 ...........................................
Predicted with 7 Factors ...........................................
Loadings Plot for Factor 8 .........................................Scores Plot for Factor 8 ...........................................
Predicted with 8 Factors ...........................................
HD2 ATR Spectrum WA913 6 .....................................
HD2 Spread on Aluminum .........................................
Model ONH Predicted with 10 Factors ............................
Model AL2 Predicted with 10 Factors .............................
Model ONH Loadings Plot for Factor 1 ............................
Model ONH Scores Plot for Factor 1 ..............................
Model ONH Loadings Plot for Factor 2 ............................
Model ONtI Scores Plot for Factor 2 ..............................
Model ONH Loadings Plot for Factor 3 ............................Model ONH Scores Plot for Factor 3 ..............................
Model ONH Loadings Plot for Factor 4 ............................
Model ONH Scores Plot for Factor 4 ..............................
Model ONH Loadings Plot for Factor 5 ............................
Model ONH Scores Plot for Factor 5 ..............................
Mirror Surface 450 .................................................
Packed Barium sulfate at 45 o ......................................
D6AC steel plate at 12o ............................................
D6AC steel plate at 45 o ............................................
D6AC steel plate at 60 o ............................................
D6AC steel plate at 70 o ............................................
Raw data of SOL01
Raw data for SOL02
Raw data of SOL03
Raw data of SOL04
Raw data for SOL05
Raw data of SOL06
Raw data of SOL07
Raw data for SOL08
Raw data of SOL09
Raw data of SOL10
Raw data for SOL11
Raw data of SOL12
........................................ ° ......
......................... ° ................... °
°, ................................... • ........
................................. °° ............
40
41
41
41
42
42
42
43
43
43
44
44
44
47
48
49
49
50
50
51
51
52
52
53
53
54
54
56
56
56
57
57
57
63
63
63
64
64
64
64
64
64
65
65
65
o°°
111
Figure 20a.
Figure 20b.
Figure 20c.
Figure 20d.
Figure 20e.
Figure 20f.
Figure 20g.
Figure 2Oh.
Figure 20i.
Figure 20j.
Figure 2Ok.
Figure 21 a.
Figure 2lb.
Figure 21 c.
Figure 21d.
Figure 22.
Figure 23a.
Figure 23b.
Figure 23c.
Figure 23d.
Figure 23e.
Figure 23£
Figure 23g.
Figure 23h.
Figure 24a.
Figure 24b.
Figure 24c.
Figure 24d.
Figure 24e.
Figure 24£
Figure 24g.
Figure 24h.
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
Predicted vs. known compositions
of Methanol using 1 factor .......
of Methanol using 2 factors ......
of Methanol using 3 factors ......
of Methanol using 4 factors ......
of Methanol using 5 factors ......
of Methanol using 6 factors .....
of Methanol using 7 factors ......
of Methanol using 8 factors ......
of Ethanol using 8 factors .......
of Propanol using 8 factors ......
Predicted vs. known compositions of Glycerol using 8 factors .......
Methanol spectrum compared to Factor 1 ...........................
Methanol spectrum compared to Factor 2 ..........................
Glycerol spectrum compared to Factor 1 ...........................
Glycerol spectrum compared to Factor 2 ...........................
Percent Variance Explained ........................................
Factor 1 of Hydroxyl Mixture Set
Factor 2 of Hydroxyl Mixture Set
Factor 3 of Hydroxyl Mixture Set
Factor 4 of Hydroxyl Mixture Set
Factor 5 of Hydroxyl Mixture Set
Factor 6 of Hydroxyl 1W_txture Set
Factor 7 of Hydroxyl Mixture Set
Factor 8 of Hydroxyl Mixture Set ..................................
Scores for Factor 1 of Hydroxyl Mixture Set .......................
Scores for Factor 2 of Hydroxyl Mixture Set .......................
Scores for Factor 3 of Hydroxyl Mixture Set .......................
Scores for Factor 4 of Hydroxyl Mixture Set .......................
Scores for Factor 5 of Hydroxyl Mixture Set .......................
Scores for Factor 6 of Hydroxyl Mixture Set .......................
Scores for Factor 7 of Hydroxyl Mixture Set .......................
Scores for Factor 8 of Hydroxyl Mixture Set .......................
................. ° .......... ° .....
............ ° .......... ° ..........
................................. °
67
67
67
68
68
68
69
69
70
70
70
71
71
72
72
73
74
74
74
75
75
75
76
76
77
77
77
78
78
78
79
79
iv
Tables
Table I.
Table II.
Table Ill.
Table IV.
Table V.
Table VI.Table VII.
Vibrational assignments for water vapors ............................. 4Samples used for UV study .......................................... 15Witness Panels Data Sets ............................................ 29
Environmental Chamber Test ........................................ 31
HD2 Grease Samples ................................................ 45
Mole Percentage of Respective Alcohol .............................. 62
Some Groups and their Wavelengths ................................ 62
V
1.0 Introduction
The aerospace manufacturing industries are currently in a state of flux with respect toenvironmental restrictions. Old (and proven) methods of manufacturing are under scrutiny;
particularly, if there are any negative connotations associated with a process or its wastes. Tounderstand processes developed many years ago, many innovations in process development are
required in order to transition between the old, known methods and new, unproven methodswhich are more environmentally sound. For this reason, new and innovative methods of surface
characterization are being used to assist in the determination of contaminants which cause weaker
bonds or debonds in solid rocket motor cases.
Technological advances in spectroscopic instrumentation have provided several new and
more potent tools for solving some of the above problems. Technologies that are applicable to
spectroscopy include higher sensitivity detectors, high speed analog-to-digital converters with
improved signal/noise ratios that allows 16, 18, and higher bits of reliable data, holographic
optical elements, optical fibers, and stronger sources of illumination. Improved computing power
at the personal computer level allows collection and processing of spectroscopic data in near realtime. Since large amounts of data can be collected in rather short periods of time, processing thedata into meaningful information and archiving data and summaries has led to the evolution of a
specialty called 'chemometrics'.
Chemometrics is a discipline which uses mathematical and statistical methods for handling,
interpreting, and predicting chemical data. Examples of chemometric methods are factor analysis
and multivariate analysis. Factor analysis is a multivariate technique for reducing complex datasets to their lowest dimensionality to yield recognizable features and/or predictions. Since there is
a strong statistical component in chemometrics, hypothesis testing foUowed by new postulates and
further testing can lead to information that is normaUy not available by direct observation.
Multivariate calibration is an approach to combining many different instrument channels in
order to reduce sdectivity problems. The foremost application of multivariate calibration today is
in Near Infra-red Spectroscopy (NIR) [1]. NIR relies upon multichannel calibrations to provide
the selectivity enhancement needed for quantitative spectroscopy in less than perfect conditions.
Examples include intact biological samples or turbid process mixtures. General benefits include
less sample preparation, higher reliability, and wider range of instnnnent application.
Optical fiber spectrometry has been around several years and has penetrated many
applications in the food and chemical process industries. With optical fibers as transmissions linesfor spectrometers, spectral information can be obtained in very difficult and sometimes remotelocations due to the flexibility of the optical fiber transmission link. Several companies currently
market such systems. The concept is expanding into areas once not available for optical
spectrometry.
2.0 ResearchObjectives
The original conceptat the beginning of the project was to demonstrate the ability of
optical fiber spectrometry to determine contamination levels on solid rocket motor cases in orderto identify surface conditions which may result in poor bonds during production. The capability
of using the spectral features to identify contaminants with other sensors which might orgy
indicate a potential contamination level provides a real enhancement to current inspection systemssuch as Optical Stimulated Electron Emission (OSEE). The optical fiber probe can easily fit into
the same scanning fixtures as the OSEE.
The initial data obtained using the Guided Wave Model 260 spectrophotometer was
primarily focused on determining spectra of potential contaminants such as I-ID2 grease, silicones,etc.. However, once we began taking data and applying multivariate analysis techniques, using a
program that can handle very large data sets, i.e. Unscrambler II, it became apparent that the
techniques also might provide a nice scientific tool for determining oxidation and chemisorptionrates under controlled conditions. As the ultimate power of the technique became recognized,
considering that the chemical system which has most frequently been studied in this work has been
water + D6AC steel, we became very interested in trying the spectroscopic techniques to solve a
broad range of problems. The complexity of the observed spectra for the D6AC + water system is
due to overlaps between the water peaks, the resulting chemisorbed species, and products ofreaction which also contain OH stretching bands. Unscrambling these spectral features, without
knowledge of the specific species involved, has proven to be a formidable task.
3.0 Optical Fiber Spectrometry Concepts
The use of silica optical fibers allows a very broad spectral region to be interrogated since
the optical transmission of silica extends from around 190 nm in the UV to around 2.5 microns in
the IR. Figure 1 shows how several spectral regions are related to wavelength, wavenumbers,
and photon energies. These relationships are significant for this work, because of the interplay
with the optical stimulation phenomena occurring in OSEE. There is no single light source or
detector which covers such a broad range. Several light sources detectors, and gratings are
available, easily interchangeable, which allows full spectral coverage with fairly limited change-out
periods for the optics.
During the course of this research we have recorded spectra throughout the range 190 nmto 2500 nm. We have used two different gratings; one with 300 lines/ram for the NIP, and
another with 1200 linesdmm for the UV. Due to the specific application being researched, that of
implementing an optical fiber spectrometer in conjunction with OSEE scanning operations, we did
not record any spectra in the visible region. Consequently we were able in general to record
spectra under ambient conditions with minimum interference from external lighting.
Due to the nature of the environment being investigated a large proportion of the spectra
were recorded under very moist or humid conditions. It is weU known that water has several
absorption bands in the NIR region. The water and carbon dioxide absorption bands make
2
W
LU
Zm
0Zm
WO.
i
0>
!
U
m
ILl
N
if')m •
I
0
0
0m 0
00
-- 00
0
-- 0
C
it
Lt_
LI
L
m
m
L
Z
D
an
°_
convential and Fourier transform spectroscopy almost impossible in the presence of water. Fiber
optic spectroscopy raises silica fibers to transport the light beam. This eliminates the long paths
through the atmosphere. Two separate published spectra for water are given in Figures 2 and 3.
Figure 2 shows the spectra published in the Handbook of Military Infra-red technology (2), which
is a compilation of atmospheric absorption features of water and carbon dioxide vapor. We
originally looked for correspondence with these absorption features in our data due to the
similarity of the chemical species involved. In addition, there was an NIP,. spectrum of water
reported by Weyer(3) that shows the per cent transmission of bulk water. Note that the
absorption are significantly different. Table I shows the spectral absorption's attributed to water
and the corresponding harmonics which would be observed in the NIR.
Table I. Vibrational assignments for water vapors
Vibration(s) Assignment Wavenumbers Wavelengths
(cm "_) (microns)
vI Sym-stretchmode 3657 2.73
v 2 Bending mode 1595 6.27
v_ +v 2 Combination band 5252 1.9
v 1 + 2V 2 Combination band 6847 1.46
2v 2 Second Harmonic 7314 1.37
v I + 2V 2 Combination band 8442 1.19
The wavelengths for the combination and overtone bands shown in Table I fall into the
spectral region in the NIR where most of the work has been performed in this research. Since
there is some discrepancy between the spectra published by Weyer and the IR Handbook, we took
our own observations to determine what the spectra looked like with the optical fiber
spectrometer. This data is shown in Figures 4. A spectrum of acetone is shown as figure 5 to
illustrate the large number of carbon hydrogen bands in the near infrared. Sharp carbon hydrogen
bonds are also observed for silicones.
Specular reflections, angle of incidence equals angle of reflection, produce spectra when
chemical species on the surface absorb light. Diffuse spectra are observed when the surface is
rough or contains small particles that result in multiple reflections from surfaces that contain
absorbing species. Absorption/reflection refers to the condition where a dielectric is covered by
an absorbing film. Light can be specularly reflected at the surface of the film or pass through the
film, with absorption, be reflected at the surface and pass a second time through the film. Since
the index of refraction of a material goes from a finite values to + _, to - _, and again to finite
values at the center of an absorption band, the band shapes appear similar to those obtained by
differentiation.
A significant aspect of the work performed deals with the absorption/reflection
phenomena associated with spectral observations from D6AC surfaces. The problem to be
4
o
%
-'- : 0 !
_" - "_ _
- 0
!
"_ 0 _ '
-- (,_)blOISSIIAIS_VIIJ-
>
<_
Sc_
el
5
° °..... ............... -
3NOI_3AO 1S_I_ O0
WN O00Z
ONV8NOIlVNIBWO_ OZH
WN og6L
(7113 WDI'O)
3NOI_3AO IS_I_ HO
WN OP_L
N N
liIII
i I I I
\ i
IC'_I
I
I°N
icO
i
t,e
J
t
(1733 WDL) ,'
3NOI_3AO ONO_3S HO .'
WN 096
I I I _ I I I
3D_BBOSBV
661M'--
Z
Z
0--
C_
v
L.
0
°_
Z
6
<f
/
/J
o o
o .oc_ o4
\
fC
J
0 _ 0
0 .0
o o
O_o J __ °°
o oo -_ oro ro
•-- _D
S£Ilql'l .qDMV[t_IOSSV
o oo oo o
S£INfl ZDNVI_}IOSgV
0m
u_nl
!;:!::iulJ_h ;J;:; ;_: ;/Jh : i ¸:;:! . ; i!3:!_- :_ _::ti ¸_ii_!__
::::::::::::::::::::::::::::::::::::::::::
. ................_
:::::::::::;:::::::;:
i!!iiii_i!iiii_il....-.-.-.......-...-:-:-:-:-:-:-:-:-:-:,.,.°.,...,.,......
:::::-27.','-i::'
.:.:.:.:.:.:.:.:.:.:......................-.................
.-...-............:-:-:-:-:-:-:-:--:._...................
ii_iiiiiiiiiiiiii!_!i_!!!..%!!_!_!!!!
,...,.,.., ,.,..,..
°_
$
overcome, particularly if an automated procedure is developed, lies in the difference observed in
spectra if the process occurs through reflection (specular or diffuse) or absorption/reflection.
Each case requires a different procedure to extract the molecular absorption features imbedded in
the signals. Figure 6 shows a cross-sectional view of the optical fiber transmission and receivercharacteristics and the optical phenomena affecting whether reflection or absorption/reflection
processes account for the observed spectra. There will more discussion of these concepts
throughout the results and conclusions sections.
4.0 Brief Description of Chemometric Principles
The chemometric concepts used most frequently in this work are basic approaches to
improving the signal-to-noise of the acquired spectra and principal component (or factor analysis)
to pick out the significant features of the spectra [1, 4-5]. The procedures performed for the
spectra presented here were procedures embedded in commercial software packages. They will beidentified during the discussion on the experimental approaches.
For improving the signal-to-noise ratio we made use of several smoothing routines. TheSavitsky-Golay method is a moving average type filter that is based on asymmetric convolutionfunction around the data point to be smoothed. Each data point is then averaged according to:
÷m
y_,_ = 2_ c./yj+JNORMi----ra
With Gaussian peaks, the signal-to-noise is improved according to the square root of the windowsize. For example, a 25 point smoothing routine provides a factor of 5 improvement in S/N.
Other routines used in this work, part of the sol,are package SpectraCalc marketed by Galactic
Inc., include the ESmooth routine which is a Maximum Likelihood filter that takes an a_priori
peakshape, such as Gaussian or Poisson, and then computes the most likely set of peaks which are
buried in the spectra. ESmooth is also a maximum entropy filter which allows it to maximize upon
the possible probability states.
Multivariate data reduction techniques are required for the following reasons[I]:
1. Lack of selectivity - No single x variable is sufficient to predict y.
2. Collinearity - Redundancy and intercorrelations between the x variables.
3. Apriori information about the nature of the data is not known.
There are several methods for reducing a data matrix into a smaller number of factors. The
algorithm ultimately used depends upon the particular sof_:ware package or characteristics of thedata set being analyzed. The first step is to combine a group of measurements into one datamatrix. The measurements may include any set of chemical or physical observations and may
include several instrumental techniques. For example, in the data matrix a row may concern a
molecular species and a column may concern a particular measurement. Factor analysis yields a
score matrix that depends on the characteristics of the molecular species and a loading matrix
which depends solely on the nature of the matrix.[5] Such a separation provides the analyst withimproved insight into the number of phenomena being observed.
A datamatrixD consistsofr rows and c columns:
dik=q C_or in matrix terms:
D = R_ C_im
Mathematical algorithms decompose DD T to give R_ and C_, that contain abstract row and
column vectors related to variance. Although these factors can be used to quantitate the
information contained in D, it is usually desirable to find a transformation matrix T such that R_Tgives a matrix, R_, related to the chemical or physical content and T1C_, gives a matrix, C_,related to the instrumental observations of real components. Since TT 1is the identity matrix, the
following relation is obtained:
D = R_I"T1C_,
Using the same notation as above, spectra that obey Beer's law can be expressed by the equation
AU l =ell Cl + e_1 C2
When the spectra of all components are available, it is possible to find T such that TIC,_ givesthe absorptivities of each component and R_T gives the concentrations of each component.When the condition exists where we have error free data, minimum noise and the spectrum of
each component is known, we can calculate T and hence the concentration of each component. Ifthe concentrations are known by an independent method of analysis, then we can calculate T 1and
find the unknown spectra. This method is frequently called the A matrix method. As noise
increases due to unknown interferents or poor signal-to-noise ratio, only approximate solutions
are available. One approach to reducing the number of unidentified factors is to analyze samples
for the analytes of interest by an independent method and to use these samples as a training set. A
Principal Component Regression or a Partial Least Square Algorithms can be used to select the
spectral features that are used to quantitate real data sets for the analytes of interest. This method
works only when the training set contains all the variables that will be encountered in future data
sets. Materials other than the chosen analytes remain unknown. R_ and C,b, are used quantkate
the analytes of interest without calculating T. If a species is present that was not in the training
set, a warning of the presence of an outlier is given.
when very little is known about the components of D, all is not lost. One can use all of the
chemistry that is known and make hypothesis to be evaluated using multivariate methods. The
postulates are reformed and the procedures repeated iteratively. A number of examples illustrating
the power of this approach occur in the literature.
10
The discussion found in the text by Martens and Naes[1] is probably the one most relevantto this work. The software package Unscrambler 1I has been developed along the conceptsderived in that text. Defining factor analysis in the data compression sense, we can write
TffiXV
where X represents the observed spectra and V is the transformation matrix which produces T,the matrix of regression factors or scores.
Spectra free from noise are not obtained in most spectral observations; therefore,
XfTP'+E
for the data set of spectra where X represents the set of observed spectra. P' is the loading
matrix which contains the regression coefficients of X on T. It is the loadings and the scores
which are presented most often in the analysis results section. E is the error or residuals matrix.
When spectral data can be related to some other experimental parameter corresponding to
the observed spectra, principal component regression analysis can be performed on this matrix.
The defining relation is now:
YffiTQ' +F
where Q' now represents the loading matrix or regression coefficients of Y on T. F represents
the residuals or unique variation in Y that is not explained by the bilinear structure of the analysis.
Partial Least Squares Regression (PLSR.) forces a common T for both the X and Y matrix.
Principal component regression (PCR) is a method that once a calibration model has beenestablished, predictions of future spectral observations can be performed.
Partial Least Squares Regression (PLSR) can reduce the impact of large, but irrelevant
X-variations in the calibration modeling by balancing the X and Y information. PLSR differs from
PCR because it uses the Y variable actively during the decomposition. However, the simultaneous
use of X and Y does provide some disadvantages relative to PCK. For instance, the PLSR needs
to utiliTe two sets of loading vectors, hence it may be more complex than PCR. Also PLSR has a
stronger tendency to overfit noisy Y-data than PCR. Neither of these conditions have been found
in the work described in this report.
11
5.0 ExperimentalApproaches
The major soft'ware packages used to support this work included SpectraCalc andUnscrambler H. Typically spectra were recorded in units of log warns and transformed into
absorbance using the relationship:
A =- log I = log I_ - log I
Much work was spent at the beginning of the research effort in determining what reference
surface to use for Ir¢ A major problem arises when a spectral feature goes negative; i.e. I <I_
then most of the matrix multiplication techniques are not applicable. Practically, a negativeabsorbance is undefined and means the reference is not valid. Several reference surfaces used in
this work include a mirror, total or specular reflecting; barium sulfate, diffuse reflecting; and metal
surfaces such as D6AC steel or aluminum. We continue to search for a reference source that
could be used for all samples. An improved reference will be required for real-time monitoring.
6.0 Results
The activities performed for this research effort provided a broad scope of experiments to
build a knowledge base upon which one could improve bonding processes in SRM's. In response
to the research objectives defined earlier, a number of spectra were recorded in both the UV andNIR regions. The spectra were typically D6AC witness panels which had been exposed to
various temperatures and humidity environments for selected periods of time. In general the
environmental exposure conditions were developed by AC Inc and SAIC as a Taguchi devised
plan to determine the effect of temperature, humidity, and time on bonding for the SRM. UAH
participated in this study using the Guided Wave 260 optical fiber spectrophotometer to record
spectra as needed.
In addition to the UV and NIR spectra presented here, UAH personnel also assembled an
OSEE scanning system at UAH and was able to get most of it going during the early part of the
contract. More emphasis was placed on spectroscopy later on in the contract; and very little was
done to produce OSEE measurements at the University. Part of the problem was that version 1of the OSEE detectors were received with the scanning apparatus and the overall sensitivity was
mediocre. The OSEE scanning systems at MSFC were much better in most functional
specifications and since that data was constantly being acquired by AC and SAIC, there was no
point to our taking the same data over.
Using multivariate analysis to better understand the spectral results was very beneficial to
building up an interpretation of the spectra obtained in view of the very difficult chemical systeminterrogated. Water from the humidity of the environments and a proposed FeOOH chemi-sorbed
oxidation species form a complex dilScult to unravel. Adding to the situation was the observation
of absorption/reflection phenomena in the observed spectra. With this complexity in
interpretation, it is very difficult to understand what chemistry is occurring without suchtechniques as PCA, PCR, and PLSR.
12
6. I Spectral Determinations of Contaminants
A number of different spectral observations were recorded in this work. At the very
beginning of the activity, identification of contaminants on SRM surfaces was the primary thrust.
Hence, the first sets of data were various contaminants on specular and diffuse surfaces.
Examples of contaminants used were Tap Magic, HD2 grease, Masking Tape Adhesive, Human
Sebum, Machine Cutting Fluid, WD-40 spray lubricant, and Lubriseal vacuum grease. The test
matrix is shown as Table H, in section 6.1.1 spectra were recorded in the UV region from 200 to
350 rim.
Brian Benson, then set up a calibration using PLSplus in SpectraCalc that indicated that sensitivity
was probably around 8 l.tg/in 2 . There was a problem, however, in obtaining consistency in the
spectral observations on D6AC steel. The spectra from D6AC steel change with time and from
location to location on the plate.
In conjunction with this activity, Morgan Wang, a graduate student in ECE at UAH
worked on developing an eddy current proximity sensor to measure the stand-off of the optical
fiber and the OSEE, very much like the arrangement of the CONSCAN at MSFC. This work was
pre-empted by the need to concentrate on making measurements of D6AC steel exposed at
various humidities and temperatures in the environmental chamber at Building 4712 at MSFC.
6.1.1 Matrix of Contaminants
UV spectra were obtained for the composition shown in Table 11. The contaminants were
dissolved in CHCI 3. Appropriate quantities of the solutions were placed on a solvent cleaned
D6AC plate using a micropipette. The additions were controlled so that each spot remained
approximately the same size as the solvent evaporated. Shown in Figure 7A is the spectrum of
the pure component. Shown in Figure 7B is the Principal Component Analysis, PCA, of the
sample matrix. As expected there are 6 factors, one for each component. This is the number of
factors that are expected in the X data if there are no interactions between component or between
any of the components and the substrate. When the components are mixed interactions, for
example, hydrogen bonding, require an additional factor for each type of interaction. When a
PLS2 analysis was done using the X and Y matrix at the same time, 16 factors are required.
Figure 7C shows the variance found for each factor. Sixteen factors were more than anticipated.
However, we have learned as this program has progressed, several factors are required to
describe the D6AC substrate. The contribution of each factor to a description of the steel
changes with time and from spot to spot on the plate. If this analysis were to be repeated, this
should be done on a freshly prepared grit cleaned surface and the time between the cleaning and
the use carefully recorded. These interactions are described in the next section.
Figures 7D - 7I show the predictions from the model versus the compositions shown in Table II.
Note that very good correlation is obtained for I-ID2, WD-40, Lubriseal, and Apiezon. The
poorest predictions were for Sebum. It should be noted that this contaminant was used in the
smallest quantities; therefore, shows the greatest error due to poor signal/noise ratio. The
greatest surprise at the time the study was made was the poor fits at 0 concentration. There is no
13
signal from the component: therefore, the worst signal/noise condition. Also recall that these
plates were ordy solvent cleaned and there is a strong gradient related to humidity on the steelsurface.
We conclude that a UV method to detect contaminants as represented by these contaminants is
feasible. The NIR should also be considered.
14
TABLE II. Samples used for UV study
Component3192
3 193
3194
3 195
3 196
3197
3 198m
3 199
3 19 10
3 19 11
3 19 12
3 1913
3 19 14
3 19 15
3 19 16
3 19 17
3 19 18
3 1919
3 19 20
3 19 21m
3 19 22
3 19 23
3 19 24
3 19 25
3 19 26
3 1927
3 19 28
3 19 29
3 19 30
3 1931m
3 1932
3 19 33
3 19 34
3 19 35
3 19 36
3 19 37
HD2
0
0
0
0
0
0
0
0
0
0
0
Tap Magic
0
0
0
0
0
0
0
0
0
0
0
0
0
0 0
0 0
0
: ¸:¸.25 'i
0
0
0
0
0
!!i!!!i':!:!:iiiii2SiiiiiI:ii:!i'i:_
0
0
0
0
0
0
0
0
: 12,5 i.
0
0
0
6.25
WD-40 Lubriseal
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0
0
0
0
0
0
00
0 0
0
0
0
• ::, ::i:¸1.20̧:: .i !
0
4O
0
0
0
0
0
0
0
0
0
Apiezon0
0
0
0
0
0
0
Sebum
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0 0
0 0
0 0
0 0
0
0 0
0 0 0
0 0
0
0 0 !: i:2.5 : :. _::
25
0
0
0
0
: I0-
0
0
0
0
0
0 0
0
15
_o5-1100-_
95-
90-
85-
80-
75-
70-
65-
:, ". ..... i I_._._RISEAId i {" _u_l't_IC :: 63"'$_k_L_" :: ....... ::
:.--"":....:.-".-...:./_......:...:....:-..:...:...:...i,..--.:.,:...:........,:...iX..-:.,:.-.:..:...:,.:..._.,......:-.:...:..:...:...:--.;_Ii : . . i ....... i ....... 6 ...... "
: : : : ..--!
, . ...---'--'""! ...... "'"'." .... i " " " : i i
" ??n n _4h o 9_h n ?_n.n "_nhn,e_ ]o o>' <3 lo lo> _-a lo 11> <:! lg t2> <:_ ]tO 1_3> <",t ]tO 14
Figure 7a. UV Spectra of Contaminants
80 ........... ! .... _,":': : " ,--'" - I • /" "
.... ?.:: " ,'" • / : t : ,,"• . , - ° . . .' . " .... • ....... , . . ._o.... / ::...... : ___ ,'" : ... :
• ./{/. . . : . . .,. .... :
• : ," : t : . :
40 .... " " " " " ;" " _ " " "
i/" .... : ' : '" :• . : ;/ .... : .." :20- " ..-'"';" i
• "" "/ E" "
n-4 . . _ . ., c.SEB'UM : i'$
fl 1 ? 3 4 'i#'t_1,tP_ ti t . ,tt,_t..t-'i _]_]I___
Figure 7b. PCA Analysis of Y matrix
100-_ % '-variance-expl ......................... : ......... ;: :: o.w_,o i_ : :• ' " _2_i_;_i _ i_..._:-_:.L._,a"i i
20- "'" "_" ........,,,_ --.'...
..:.-'=-:.-.,-.=.==:..-.--¢-"_..... ...i.............. ::.........." i :: i _:_UM i0-t " " _..ors
.... 6N ..=.i .h _. 64 _ 6q 1" 1'7, 1"4 Ik
Figure 7c. PLSR of Y and X (UV data)
16
50-
40_
30-
20-
10_
0-
-10-
i t_Rl_g E_ Y-vat-
Pr, ldieted (with 16 factom) ....... : ....... : ....... : ....... :_
; :
' ............... : ..................... Me_uiz_t
Figure 7d. Regression Analysis for HD2
80- Prt
60-
40-
20-
0-
-20-
GR1E_ E. f) Y-vat-*
|
.... : .... : .... : .... : .... : .... : .... : " " Measured1 N 9n "gN 4N _0 _N "71"1 RN
WD40
Figure 7e. Regression Analysis for WD40
40- Pn
30-
20"
10-
0-
-10"
dieted(with 16 factors) • : .... : .... : .... : .... : .... : .....
t . .......
..................................... Measured
_lal_.gi _. Y-vat-- LI.I_RT
Figure 7f. Regression Analysis for Lubrisea/
17
50-1Pri_dicted (with 16 factom) ....... : ....... : ...............
40- i ....... : ....... ! ....... i " ! ii i " . . i ..... i ....... i
O- {If " : : : :-10- ....... _ ....... : ....... : ............. Measured
h lh " 9.h " _ 4b _ht".1_1_._ 1R. T--vat z APTI_._.ON'
Figure 7g. Regression Analysis for Apiezon
15.0- Pr,_dieted (with 16 factors) ....... i ....... i ..... _ i _412.5 -
10.0 -
7.5-
5.0- i
2.5- i
-5.0 -
r ..... • : : : :
If i ........... i ....... i ....... :
3 ....... : ....... i ....... ! ....... i ..... Measured
lil f_ 7h ?,sq_L_CTC
Figure 7h. Regression Analysis for TapMagic
2.0-, ....... ....
1.0- Io.5-_-...... ! ....... i..._... ::....... i ....... !
O- _ i : : : : :
-0.5- _.,L ....... ! ....... : ....... : ....... : ....... i
-1.0- 15 ....... i ....... i ....... i ....... : ..... Measured
1
Figure 7i. Regression Analysis for Sebum
18
6.1.2 Choosing the Reference Spectra
When working with spectrophotometric data Absorption Units are preferred.
ALl =- I/I_ = Log(I_.) - Log(I)
The Guided Wave Model 260 is a single beam spectrophotometer. Therefore it is necessary to
record the spectrum of a reference and store it in the memory to be used when the sample data is
taken. Choice of the reference sample is very important. For many purposes a reference is
recorded, the spectrum of the sample taken, and then the reference again checked. For the
experiment described in Section 6.1.1, a spot on the blank plate was chosen for the reference and
the sample data referenced to this spectrum and the data recorded as Percent Reflectance (%R).When the laboratory initially received the D6AC panels there was not a comparable procedure.
The entire panel had been treated. A search was made for an appropriate reference. Frontsurface mirrors, pressed BaSO4, thin layer chromatographic plates, and Kodak Color References
were tried. None of these were satisfactory for this project. The Guided Wave Model 260 can
write to disk a log(watts) file. Therefore, the data has been archived as log(Watts). These files
can be used to try various data reduction methods as other reference samples become available.
The spectral differences, when the same spot is observed as a function of time, are usually small
(0.000 to 0.03 AU). If the reference is not a perfect match and we subtract the sample spectrum
from the reference spectrum, negative values are obtained. A negative state for AU is undefined.
The condition can only be defined as the sample reflecting more light than the reference. If aspecular reference is used AU values greater than 4 are obtained. If a true value, then 99.999
percent of the light is being absorbed.
For the D6AC data which was recorded as a function of time and included in this report the first
spectrum obtained was used as the reference and the later values subtracted from it. This gives aconsistent set of data for a given condition. Comparison between tests where conditions differ areless reliable.
As indicated above many of the values were buried in a noisy signal. Therefore after the
subtraction was made the noise was smoothed using a maximum likely hood- maximum entropy
routine (6). For these studies a Gaussian shape peak and normally distributed noise was assumed.The effect of the smoothing is shown as Figure 8.
The search for a standard that can be used for all types of samples continues. We are also
investigating other approaches to data reduction. Any time a mathematical operation is
performed there is a chance of adding noise (data that is a function of the method rather than the
sample). For many purposes watts would be better than log(watts). The values for reflected
energy from steel and aluminum for the Guided Wave Model 260 is 10 '° to 10"'. The antilog is a
smaller number than the PC can store in a buffer. An approach that we are currently trying is to
add 9 or 10 to the data to multiply the data by 109 or 10 '° and then take the antilog. The new
spectra are suitable for spectral subtraction and similar procedures. Multivariate analysis
19
procedures can be used on the data without further processing. This procedure may be applicable
as we move toward automated inspection with automatic data reduction and display.
6.2 Chemometric Determinations
Several types of information can be derived from accumulated spectroscopic observations on the
sets of witness panels that have been processed by a standard procedure. The initial observations
were related to possible correlations between spectral data and OSEE data on witness panels that
were placed in an environmental chamber at MFSC and maintained at different temperatures and
relative humidities for various times up to 28 days. The conditions were meant to simulate
potential working environments for SRM manufacturing to determine whether some of these
conditions would affect bonding parameters for the chemical system used in the SRM. The test
specimens and environmental conditions are given in Table rrt. PCA analyses were made using 4
to 8 spots distributed uniformly on the test specimens. As we attempted to obtain reproducible
results from the analyses we began to realize that the reactions do not terminate when the samples
are removed from the environmental chamber and stored in nylon bags containing nitrogen. The
time between removal from the environmental chamber and the NIX observations made at UAH
varied from a few hours to several weeks. A cluster analysis indicated that there was a difference
between the 28 day sample and the samples stored for shorter periods. The reference problem
discussed in the last section was discovered. Since log watts records had been archived, several
spots on the plates were tried as reference spots. The magnitude of the differences within a panel
are shown in Figure 9.
The surfaces were also changing as the samples were exposed to the UAH laboratory
environment. We were convinced that the spectra represented absorption bands for water
physically and chemisorbed, and the OH bands of reaction products. We also were confident that
the Guided Wave Model 260 Spectrophotometer was sensitive enough, using the probe
containing 9 fibers to iUuminat¢ the surface and 10 fibers to observe the surface, to provide real
time data. The Guided Wave Model 260 spectrophotometer was set up on two occasions at the
MFSC chamber. These experiments were labeled E4 and HIT. The test conditions are given in
Table IV. A third series of data was obtained in a UAH laboratory. A small laboratory was used
for the experiment. This laboratory has no outside windows or doors. The environment was
found to be reasonably constant over a 2 week period. The test conditions are also given in Table
IV.
All of the real time data was processed by subtracting the values for a sample from the initial
sample to convert to absorbance units, using ESMOOTH and when necessary correcting the base
line to 0 (offset). The spectra that are obtained do not have a baseline reference and the peaks
appear to overlap. A common method to deal with these conditions is to convert to the second
derivative. Many peaks were observed. The soitware available at the time was not suitable for
multivariate analyses on the derivative spectra. Therefore, the absorption spectra were used with
the PCA algorithm found in UNSCRAMBLER II. The 78 spectra in the E4 set were considered
as one set. Six to eight factors were required to model the data. Considering the large number
of possible compounds this is considered a small number. In an attempt to determine the time
sequence of the appearing and disappearing spectral lines the data was divided into several sets.
20
For the first attempt a factor analysis of spectra 1 through 10 was used followed by analysis of
spectra 5 - 15 etc. The individual sets were often modeled with one factor and seldom showedmore than 3 factors were required. The factors found in these sets were combinations of the
factors obtained on the entire set. This procedure was used in an attempt to see if reactions wererepeating in a sequence. As we have accumulated data it is now believed that the segments weretoo small, that is the reactions are slow. The model would be consistent with a long induction
period before a reaction begins and a fast reaction following initiation. A significant variant is the
absorption and deabsorption of water vapor. A third procedure was tried. Samples 1 - 15 wereused, then 1 - 25 etc. The analyses followed the scores and loadings of the total set. However,
we did get some indications where reactions started. A companion study that starts from the back
and moves forward was not attempted as data from the second series was becoming available. As
described above, PLS algorithms use the X matrix with a Y matrix. A Y matrix that used the
sample number was used with the 78 spectrum matrix. It was necessary to remove spectrum 67from the data as an outlier. It was then found that the sequence was predicted correctly. The
exposure time in minutes was used to replace the sample number in the Y matrix. The result wassmoother curves.
Since we are modeling data related to compounds with unknown identity, data reduction requiresiterative attack. Both time and OSEE values were available for the HIT series. Now we have
two columns of values for the Y matrix. Therefore, the PLS2 algorithm in UNSKAMBLER II
was used. Shown in Figures 10a and b are the predicted values for time and OSEE vs observed
values using 10 factors. The predicted values for time are better than the predicted values for
OSEE. The largest differences in the OSEE predicted and observed values are at the beginning of
the sequence when the OSEE values are changing rapidly and at the observations following power
interruptions.
Model SO C constructed using the UAH Lab data was used to predict the values for the
E4 and HIT sets. The model predicted the correct sequence; however, the times are different
from the observed. The 70o runs predicts high values for the HIT set. This is what would be
expected if endothermic reactions are taking place. Less time would be required to reach a givenstate at the higher temperature. The three rates are composed in Figure (11).
A number of attempts were made to extract kinetic data for oxidation of the D6AC.Unfortunately the data for E4 and HIT data sets showed sutficient oscillatory behavior in thefactors over time that we are not confident that the kinetics of the oxidation species are obtainable
from the data. We also are wary of environmental control in these tests which might affect the
oxidation and hydrolysis of the D6AC steel.
21
-|
__i__'_-
-,,=_
-'-2
0 0 0
0
S£INIFI _t;3NIVI3_IOSI3V
22
00
o4
00
O3
0Z
Z
o Z _0 _.10
00o,1
rj
o3
&
6.3 Discussions Related to Two Models
Two data sets are discussed in this section to illustrate the methods which are used. Shown in
Figures 1 la through 1 lj is results of an early model, NAg A. The data is taken from the E4 set.
Loadings are plotted at the top of the page and scores at the bottom of the page. Wavelength -
1000 nm is the abscissa for the loadings. The loadings are selected features of the spectra. The
scores, representative of amount of a species, are plotted vs sample number. Note that factor 1 is
primarily for sample 67. This was a spectrum recorded aider a power outage. Factor 1 is so
dependent on a single factor that we would repeat the procedures and obtain a new model. This
was done. However, luther discussionofModel NAg A is profitable.
We must remember that we are working with unknown compounds and their spectra. As we have
continued our investigations looking at additionfl models we are beginning to interpret this early
data. The three predominant peaks in Factor 1 the loading appear to be first derivatives. The
shape of these three peaks suggest that they result from refiection/absorbance phenomena i.e. the
beam passes through a film is reflected at the steel surface and makes a second pass through the
film before reaching the spectrophotometer. A current hypothesis is that the composition of the
film is some form of water. The peak at 1000 (2000 nm) has the reverse shape to the peak at
approximately 1450 and 1100 nm. If these peaks are related to I-I20 the thickness of the peaks
near 1450 and 1100 nm are greater than the thickness on the reference (sample 1). The peak at
2000 nm is less than the reference. This is the usual pattern in most of the spectra. Recently we
have recorded spectra in the laboratory in which aLl three of these peaks were increasing. These
changes are related to changes in teh relative humidity. Looking at Factor 2, we find that this
factor looks similar to the first factor but peaks around 2400 nm are important to the last 10
samples. These could represent a hydroxyl containing species. Since it is late in the run these
species may be more stable. Note that there is no indication that the beam has passed through a
film. Note that the same peaks are involved and that sample 67 is a heavy contributor. Factor 4
loadings show strong bands near 1450 and 1900 nm. The scores indicate that these absorbtion
peaks are important throughout the run. Note that the peaks appear to contain more then one
component. We have noticed that the ratio of these peaks can be different from sample to sample.
We offer no interpretation for Factor 5. It may be related to teh metal surface.
Model SO version C is one of several models made for the run that was done in the UAH
laboratory. The Model is described in Figures 13a through 13x. Looking first at the predicted vs
observed plot at the bottom of the page, we find that one factor describes a lot of the variance and
has a general upward trend. However with only one factor there are a number of peaks and
valleys. The spectra represented in the loading all show intensities less than the reference. The
peak at 2000 nm is easily observed but does not appear to have sufficient thickness that the peak
has the shape of a derivative. Note that the peaks near 1850 nm are almost resolved. Turning to
Factor 2 prediction vs observed curve at the bottom of the page we see that some of the valleys
have been filled in. The scores plot indicates the quantitative aspects. The scores curve shows
that Factor 2 was important for the 35th through the 60th spectra. The peak at 2000 nm now
appears to be a derivative indicating a thicker film. In addition to the peaks around 1900 there are
feature around 1400 nm. where OH bands would be expected. Turning to Factor 3, we find that
the predicted vs observed curve continues to straighten out. The bands at 2000 nm, 1450 rim, and
23
00r-.
I
/
C',4od
I
(
-,,.,co0d
I
C)(D
c'q
00T--
0.4
0'_I
S££V_ OO_I
*tm_
rd'a
=C)
k,
=
.e_ r,.. r,_oZ _
,- r._
O
Qr_
o
t-N•-- Q,I
opnl
24
1750-
1500-
1250-
1000-
750-
500"
250-
i !
i........................0 - Measured
6 .... 250 .... 560 .... 750 .... 10_"' '1256'' "15'00"" '17'50"" :20_
HIT E, ¥-var: MINUTES
Figure 10a. Observed vs Predicted times for samples in HIT series
420-
390-
360-
330-
300-
270-
Pr.edict._d.(with .10.facto.rs) ......................... _ " . .
ii ii ii ii ii 4o_i9_ ii198 7./6../_ 4 iID
! ..... :. ..32: ..... : ..... : ..... i ..... : ......
Measured
2")5 .... 360 .... 315 .... 350 .... 3")5 .... 460 .... 415 .... 450
HIT E, Y-var: OSEE
Figure 10b. Observed OSEE values vs Predicted values in HIT series
25
"L
• . • _ "=. _ • - . r-._ V
.............. _= ....i _ ............. _: =-
.... _. _p,_ • . ,_
.... _. - ,..,, • . .
.... _ _-_ • . . _,_ __
.... _ i _. • - - "_
, , , , _ _ ==
26
around 1100 nm appear to represent spectra of thin films. Note that there are two peaks in the
1100 region. As we continue to add factors the predicted vs the observed curve improves. It is
difficult to interpret the loadings plot. The magnitudes are decreasing and noise is observed. Theobserved spectral features are consistent with the NAg A model.
6.4 Observations Related to HD2 on D6AC Panels
An investigation related to methods of covering steel panels uniformly with a known weight of
HD2 was running concurrent with the environmental test chamber experiments. While the Guided
Wave Model 260 Spectrophotometer was at MSFC, reflectance measurements were recorded in
log(watts). Some of the first measurements were made on panels that could be identified as beingnonuniform by visual inspection. For the first group of samples, a reference was taken before thepanel was coated. For the second group of samples, the panels had been coated before thereflectance measurements were made. A reference was taken of a solvent cleaned spot. Including
blanks a total of 48 spectra were taken. If a spectrum of HD2 on the substrate was available thiswouid be a simple calibration. We have worked with this set of spectra as we investigate better
methods of gaining the maximum information from a data set
The 48 spectra have atl of the features and problems identified in previous discussions. There is
some evidence that the panels had the film that has been tentatively identified with waterabsorption and deabsorption before the coating operation began. Also the coating procedure may
be adding water in addition to the trapped water. The first attempts to visuatly identify peaks that
were increasing as the thickness increased was unsuccessful. An ATR spectrum was recorded.This spectrum is shown in Figure 14. More recently, we have obtained a reflection spectrum ofI-ID2. This film was sufficiently thick that only reflections from the surface were obtained. This
spectrum is shown as Figure 15. A series of models were made systematically reducing the
number of spectra in the calibration set until a consistent set was obtained. There is significant
scatter in the data. Some of the variance is probably related to non uniformity resulting from
separation of the chemical components.
Rather than report the data related to the models discussed above, some recent work related to adifferent method of handling the data is reported. The reflection curves are recorded in
log(watts). The procedure for reducing the data to AU has been described. Since adding orsubtracting logarithms is equivalent to multiplication and division the original data cannot beaveraged or smoothed prior to conversion. An additional objective is to reduce the number of
mathematical operations since each operation can add noise (data not related to the sample).Since the values are approximately 10"1°,if the ant/log is taken the values are too small for the PC
buffers. Spectra appear on the screen with 10 zero's for the absorbance units and values can notbe saved to disk files. We add 10 (equivalent to multiplying by 10l° and then take the antilog.
Using this procedure, we obtain a spectrum in watts. These spectra can be averaged, subtractedfrom each other, or a ratio obtained. Any of the multivariate analyses can be made directly on the
spectra. During this time period Dr. Arendale had a Perstorp SCL spectrophotometer on loan.This instrument has less than 20 microAU noise level. Therefore, more digits must be
accommodated. We used the SCL software to process the data. Since there is significant
differences in the signal level for the various samples, each sample is normalized to have a value of
27
1.0. This results in the elimination of the mean as Factor 1. A prediction vs observed graph is
shown as Figure 16a. Model ONH B was obtained a_er removing outliers (spectra 13 - 14, 24,26 - 28, and 30). Sample numbers and spectrum identifications are given in Table V. The
prediction for all samples is shown on Figure 16b. The scores and loadings for Model ONI-I B are
shown in Figures 17a through 17m. Factor 1 shows features that can be identified in Figure 15,The water and OH lines begin to dondnate in the other factors. Some of the variance is probably
due to specific compounds that are used in the I-ID2 formulation.
28
TABLE HI. WITNESS PANELS DATA SETS
SAMPI,E TITLE IN UNSC.
F.,6B1 (TOP
ORIG. SC NAMEOSE6B_I
2 E6B2 (TOP RIGHT) OSE6B_2
3 E6B3 (BOTTOM RIGHT) OSE6B_3
4 E6B4 (BOTTOM LEFT) OSE6B_4
ESA1 (TOP LEFD
ESA2 (TOP RIGHT)
ESA3 (BOTTOM RIGHT)
ESA4 (BOT'I'OM LEFT)
ESA5(lOPLE_E8A6 (TOP RIGHT)
ESA7 (BOT'I'OM RIGHT)
ESA8 (BOTTOM LEFT)0E830A10 (TOP LEFT)
7
8
9
10
11
12
13
OSESA_I
OSESA_2
OSESA_3
OSESA_4
OSESA_5
OSE8A_6
OSESA_7
OSESA_S
SE830A10
14 0E830AI 1 (TOP RIGHT) SE830A11
15 0E830A12 (CENTER) SE830A12
16 0E830A13 (TOP LEFT) SE830A13
17 SE830A140E830AI4 (BOTTOM
0E830AI5 (BOTTOM
RIGHT)
OS7221 (TOP LEFT)
OS7222 (TOP RIGHT)OS7223 (BOTTOM
RIGHT)OS7224 (BOTTOM LEFT)
Ios7225(CENTER)
OS7231 (TOP RIGHT)
OS7232 (TOP LEFI)
OS7233 (BOTTOM LEFT)
OS7234 (BOTTOM
RIGHT)
OS7236 (CENTER)
OS7237 (BOTTOM LEFT)
OS7238 (BOTTOM
RIGHT)
OS7239 (TOP RIGHT)
OS72310 (TOP LEFT)
18
_19
20
21
22
23
24
25
26
27
28
29
30
31
32
SE830A15
OS7 22 1
OS7 22_2
OS7_22_3
OS7 22 4
OS7_22_5
OS7_23_1
OS7 23 2
OS7 23 3
OS7_23_4
OS7 23 6
OS7_23_7
OS7 23_8
OS7 23 9
OS7 23 10
DESCRIPTION
!__i_i_!i!ii!i!i!iiiiiiiii
!!! i! i iiiiiiiiii!i!iiiiii!iiiili!iiiiiii
__i_iiiiiiiii!iiii_ii!ii_i_iiiii!iiii!!iiiiiiiiiiiiiiiii_iiiiiiiiiii_!_!ii_!i iiiii!iiiiiiiiiiiiii
i__ !_i::O_i!ii!::i!!ii!iiiii
29
33
34
35
i36
37
IOS7246 (CENTER)
0S7247 (BOTTOM LEFT)
OS7248 (BOTTOM
RIGI-rl')
OS7249 (TOP RIGHT)
0S72410 (TOP LEFT)
OS7_24_6
OS7 24 7
OS7_24_8
OS7_24_9
07 24 10
i__i_!iiiii!ii_iiiiiiiiiiiiiiiiii!i!ii_ii_!i_i_i!iii_iiiiiiiiiiiiiiiiiiiii_iiiii_i!_@ii__!iiiiiiiiiiiiii!iii!ii!i!ii!!i!!iiii
30
TABLE IV. ENVIRONMENTAL CHAMBER TEST
Spectra Number
E4 1-50w
E4 51 -78
E4 Data Set
(50°F/60% RH)
Date Spectra was taken
August 6, 1992
August 6 - 12, 1992
Time Span Between each
Spectra
One Hour
Four Hours
Spectra Number
HIT 1 - 17
HIT 18 - 44
High Temperatuare (HIT) Data Set
(100°F/60% RE D
Date Spectra was taken
September 16, 1992
September 16 - 17, 1992
Time Span Between each
Spectra
15 Minutes
One Hour
Spectra Number
D6AC 1 - 17
D6AC 18 - 40
D6AC 41 - 115
D6AC Data Set
Average Temp. 73.5
Average Humidity 36.7
Date Spectra was taken
September 28, 1992
September 28 - 29, 1992
" September 29 - August 12,1992
Time Span Between each
Spectra
15 Minutes
One Hour
Four Hours
31
Loadings .........0.06- _ .............................I
0.05-
0.04-
0.03 -
0.02-
0.01 -
O_
6 3t_O 600 960 1200 1500
NAR A, factor.
Figure 12a. Model NAR Loadings Plot for Factor 1
0.8-
0.6-
0.4-
0.2-
_
-0.2-
Scores .......................................
1(_
J
)
Objects6 ......... lb ....... 2'o ....... '3'0'....... 4b' ....... 5b' ....... '6'0'....... 'q'o....... 8'o
NAR A, fac(exgl ) .
Figure 12b. Model NAR Scores Plot for Factor 1
32
0.08 -
0.06 -
0.04-
0.02-
.
-0.02-
-0.04 -
Load.ings ......................................
i .....
X-variables
() 360 660 900 12'00 15'00
NAR A, factor.
Figure 12c. Model NAR Loadings Plot for Factor 2
0.05
0.04
0.03
0.02
0.01-
0-
-0.01 -
-0.02 -
-0.03 -
Scores
2(
ObjectsI ......... I ......... I ......... I ......... I ......... I ......... I ......... I ........ '' I
0 10 20 30 40 50 60 70 80
NAR A, fac(expl).
Figure 12d. Model NAR Scores Plot for Factor 2
33
0.075-
0.050-
0.025-
O.
-0.025-
-0.050-
-0.075 -
-0.100-
Loadings.................................... _.
i
V
X-variables
6 360 660 9()0 12'00 15'00
NAR A, factor.
Figure 12e. Model NAR Loadings Plot for Factor 3
0.06"
0.05-
0.04-
0.03
0.02-
0.01-
0-
-0.01-
-0.02-
Scores
3(
I
Objects6 ......... 1'6 ....... '20' ....... 30 ....... 40 ....... 50 ....... 6'() ....... ;7'6 ....... 80
NAR A, fac(exp1).
Figure 12f. Model NAR Scores Plot for Factor 3
34
0.08-
0.06-
0.04-
0.02-
.
-0.02 -_
-0.04 -
Load.ings.
........! i.......... I iiii!
i..................................... X-vafiabies
6 3_ 66o 9()0 12'00 15'00
NAR A, factor.
Figure 12g. Model NAR Loadings Plot for Factor 4
0.03 -
0.02 -
0.01-
_
-0.01
-0.02
Scores ........................................
....iI ......... I ......... [ ......... I ......... I ......... I ......... I ......... I ......... I
0 10 20 30 40 50 60 70 80
NAR A, fac(expl).
Figure 12h. Model NAR Scores Plot for Factor 4
35
O.075"
0.050-
0.025 -
-
-0.025 -
-0.050-
-0.075 -
Loadings ......................................
f
I
/ .............................. )b : ....... : ....... : ....... : X-_a_iibi_
d 3_0 6(10 9_ 12'00 15'00
NAR A, factor.
Figure 12i. Model NAR Loadings Plot for Factor 5
0.015-
0.010-
0.005 -
0-
-0.005-
-0.010-
-0.015
-0.020-
-0.025
-0.030-
Scores .............................
5_
..................................... Ot;i,ct_I ......... I ......... I ......... I ......... I ......... I ......... I ......... I ....... /' I
0 10 20 30 40 50 60 70 80
NAR A, fac(exp1).
Figure 12j. Model NAR Scores Plot for Factor 5
36
0.02584-
0.02582-
0.02580-
0.02578-
0.02576 -
0.02574-
0.02572-
0.02570-
0.02568-
SO
I.x ,adiags
iiiiii ii !iii iiiiii! !iii!i!i.............. ',...... l", ....... i ....... _....... : ....... : ....... : ....... :''" X-variables
0 1 "_'1_ 1 gig) 19'00 2_'fiO 251")0C. factor.
Figure 13a. Loadings Plot for Factor 1
80-
60-
40-
20-
0-
-20-
-40-
-60-
A/
Sc3res ................ : ..... : ..... : " + " " " i f"_ _ "
11......... 1'I ......... ?.'2......... .3'2......... 4'2......... ';i'2........ 6'2 '71"IC. factexDl_.
Figure 13b. Scores Plot for Factor I
8000- Pr!:dicted (wilh 1 helots) • . • : ............. : ...... 6364656667686_
6000-
4000-
2000 -
_
so
5051............. : ............. : ..... 62
• 61
• 4849 " 60
: ....... 52 • "5556-75859'54243 :
.2_,,,a_2 . . ._- • • 4545 47 54 ::41... :: "46 ....... 53.... :: ............ "•",t_t
fg0 "4"JJ_9 40 • . .
_8 " "411_" Measu,ed0 "+6nO 6600 0600
C. ¥-var: MINUTES
Figure 13c. Predicted with 1 Factor
37
0.075-
0.050-
0.025-
0-
-0.025-
-0.050-
-0.075-
-0.100-
SO
i -c. _0,,,-tor. 1.3'0o , _'nn
_adings .... : ....... : ....... : ....... : ....... :
X varmbles
Figure 13d. Loadings Plot for Factor 2
2.0- Sc
1.5-
1.0-
0.5-
0-
-0.5 -
-1.0-
-1.5-
-2.0-
so
OreS .... : ..... : ..... : ..... : ..... : ..... : ..... ;
..... i ............... i ...... i, : : : i : :, : : : : : : : : obi_0 fac (ex 11oI'_ 22 32 42 _;9. 6:2 70C. °
Figure 13e. Scores Plot for Factor 2
8000 -
6000-
4000-
2000-
0-
-2000-
so
Pr,;dicted (with .2 factors) • • . :: ............. :: ...... 63 ' ' 666"/6869.
............. ii ..... 505152 .... i!. 59606162.. 64.65 •: 49 55565.758
.......... 424344454_i 4748 ..... 5354 ...... ......... i4041 : : i
• . .
...................................... Measu_d
C. Y-var: MINUTES
Figure 13f. Predicted with 2 Factors
38
0.100-
0.075 -
0.050-
0.025-
0-
-0.025 -
-0.050-
-0.075 -
_N (2. _ctor.
La ading_ .... : ....... : ....... : ....... : ....... :
ii :iiiiii:i!!!i!..... : i.......ii i ilil : i
b es
13'fM-I IgO0 10'flO 9._00 _S'O0
Figure 13g. Loadings Plot for Factor 3
2.0- Sc
1.5-
1.0-
0.5-
0-
-0.5-
-1.0-
-1.5-
SO
3rcs .... : ..... : ..... : ..... : ..... : ............
..... : ................................. Objects|
O ex 11D1_ 22 32 42 52 ...... dg.......... 7'0C. fat:: t
Figure 13h. Scores Plot for Factor 3
10000-
8000 -
6000-
4000 -
2000-
0-
SO
Pn;dicted (with 3 factors) . • • : ............. : ........ 68 "ii
i ............. :: ............. .::... 61.62 . . . 6667 . 69
! • " 60 6364! ............. : ............. : ........ 65
! :: 505152 545556575859............. i .... 49 .... 53... i ............. i
434445464748 "
!f9404142iw_ Measmed
O 3600 6600 o6nnC. Y-var: MINUTES
Figure 13i. Predicted with 3 Factors
39
0.100-
0.075-
0.050-
0.025-
0-
-0.025-
-0.050-
-0.075-
SO (2.
l.a adings, .... : ....... : ....... : ....... : ........
....... i ....... _....... ! 1.... i ....... i
,
t0aet or. 13'00 1gO0 1q'OO _'OO _(OO
Figure 13j. Loadings Plot for Factor 4
0.75 -
0.50-
0.25 -
0-
-0.25 -
-0.50-
-0.75 -
-1.00-
so
S,_:_res_.... i ..... ! ..... i ..... i ..... i ..... i A " " i
0 exD 111'_ 22 32 42 _2 62 7012. fae(
Figure 13k. Scores Plot for Factor 4
10000-
8000 -
6000-
4000 -
2000-
0-
SO Co
Prl ,_dicted (with 4 factors) . • • : ............. : ..............
6869."..... i ......... : 6667.
: :: 61_626364
• 60 65
............. i ..... 5051' " 545556! 73859 ........... i
• 49 5253
4344_45464748
_11__9404142 ' i ............. ii .......... Mi_u iied
¥-var: MINUTES
Figure 131. Predicted with 4 Factors
40
so
Figure 13m. Loadings Plot for Factor 5
1.00-
0.75-
0.50-
0.25-
0-
-0.25-
-0.50-
-0.75-
SO C
Sc DI_S .... : ..... : ..... : ..... : ..... : ..... ; .....
i
....................................... Objects
"f ][_{I ac fexDl'l " ')_'2 _h ,h _h 6h "Jr}
Figure 13n. Scores Plot for Factor 5
9000- Pr
6000-,
3000-
_
SO
,dicted(with5 factors).., i ............. ! ........ 66676869.:
i i 606162636465 i59............. ............. ...........
::: 50515253545556:_'i 58 :::49
........... 43.4d45. 4647 .......... :: ............. i• 48 " "
33_9 404142 i i •_=_,- Measut:ed
C. Y-var: MINUTES
Figure 13o. Predicted with 5 Factors
41
0.08-0.06-0.04-
0.02-
0-
-0.01 -
-0.04 -
-0.05 -
-0.08 -
SO C.
Lc adings ..... : ....... : ....... : ....... : ....... .
: _ ....... • ....... i " i i
fi 13'o0 1ann 1o7_ _gan _'nofactor.
Figure 13p. Loadings Plot for Factor 6
0.50-
0.25 -
0-
-0.25 -
-0.50-
SO
So,tog.... i ..... i ..... i ..... i .... 'i .... i ili
DAiA........... i t i
..... !........... i..... i it' o ,.io0 11 2_ "_2 42 52 62 70
C. facfexDl_.
Figure 13q. Scores Plot for Factor 6
9000- Pn
6000-
3000-
--
_dicted (with 6 _om) ......................... 6869
• " 64656667 i6263
! ! _.96061 i............. ! ............. _758 ........... i
i 50515253545556i i49
4_45464748 i :........... 43: ............. : ............. !
_,_,_ 9404142 ." :. .:Me_u_d
fi _6oo elnhn 9oonSO C. ¥-var : MINUTES
Figure 13r. Predicted with 6 Factors
42
0.075-
0.050-
0.025
0-
-0.025-
-0.050_
-0.075-
-0.100- ....... : ........................... X-vadabJes
_aetor 13'00 Ifi'00 19'00 2_'00 2(On
Figure 13s. Loadings Plot for Factor 7
1.00-
0.75 -
0.50-
0.25 -
0-
-0.25 -
-0.50-
-0.75 -
SO C.
Sc :)res .... _ ..... : ..... : ..... : ..... : ............
: i i!! !!!!!ii ii
iii ii iii iii ii iii i iiiii ob_o_, ......... , ......... , ......... , ......
Figure 13t. Scores Plot for Factor 7
so
9000- Pr,
6000-
3000-
.
'dieted (with 7 factors) • • . : ............. -: ....... 64 66 _-_76869"• 65: : 62 •
i 57 596061 63 ::
............. :: ..... _,,51" "53545556 i 58 ............ ii ou 52 :: 49 : :
........... _134_.'45"4"64748 ......... i ............. .
9404142 i " •w--v Measm'ed
O 30'00 6_o 9600C. Y-var: MINUTES
Figure 13u. Predicted with 7 Factors
43
0.100-0.075-0.050-0.025-
0--0.025-
-0.050--0.075--0.100-
St) C.
Lc adings .... : ....... : ....... : ....... : ..... :
_actor. 1_'00 1600 1900 22'00 _'00
Figure 13v. Loadings Plot for Factor 8
0.6"
0.4-
0.2-
0-
-0.2-
-0.4-
-0.6-
-0.8-
SO
Sc
{_ ................................................. _ ......... _ .........
C°
0,es.... i ..... i ..... i ..... i ..... i ..... i ..... f
:i '< i:vvV ....viv
..... ! ..... : ........... : ..... : ............
........... : ........... : ..... : ..... :' Obieetsm
11 22 32 42fac f exDl _.
Figure 13w. Scores Plot for Factor 8
so
10000-
8000-
6000-
4000-
2000-
0-
Pr ,_dicted(with8faetors) • • . : ............. : ............. :69
.......................... i ............ :i :: 606162636465666768::
• i ............. 57-^59 ................. : 51 53545556 i a_S ..... !
: 50 52 :
........... 43_.45464_/4849 ........ i ............. i[
9404142' . . .w -,-,, Measu,'ed
C. Y-var: MINUTES
Figure 13x. Predicted with 8 Factors
44
TABLE V. HD2 Grease Samples
Sample Number1
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
Sample Identification8181
8182
8183
8184
8185
8186
824 1
8242
8243
8244
824 5
8246
926
927
828 7
8288
8289
8 28 10
8 28 11
921
922
923
924m
925
8247
8248
8 24 9
8 24 10
8 24 11
931
932
933
934
8187
8188
8189
8 18 10
8 18 11m --
935
936
937
MG/FT 2
0
0
0
0
0
0
0
0
0
0
10
10
10
10
10
10
10
10
10
20
20
20
20
20
20
20
20
45
42 9 4 1 2043 9 4 2 2044 9 4 3 3045 9 4 4 3046 9 4 5 3047 9 4 6 3048 9 4 7 30
46
IJ
}\
-%
c'q0
S£INgl XDNV_IOSgV
00
cxl
0
0
0
O_
zm_
_ z _'tO _
00
47
0c.C)
J
00
0L_
0 0CO
C)00
00
E_
0Z
Z
ZI---I
48
-
30_ Predi.cted (with 10 fa.ctors) ......................
'.............. i ; )i P21 : : : " "
-5-Measured
lb .... 1'5 2'0 2'5 .... 3'0ONH B t Y-var: -,. MG/FT2
Figure 16a. Model ONH Predicted with 10 Factors
30-
25-
20-
15-
10-
5_
-
-5-
, ,.... iiiiii
28
!
Reference
6 _ lb 1'5 2b :_5 3b
AL20NHB, Y-var: MG/FF2
Figure 16b. Model AL2 Predicted with 10 Factors
49
0.08 -
0.06 -
0.04-
0.02-
-
-0.02-
-0.04-
Lc ,adings ................
0 1300 1600 1900 2200 2500
B factor.
Figure 17a. Model ONH Loadings Plot for Factor 1
8- S_,rcs-A._ ......................./
6-_ .....................................
i
4 =
0" r-
-6-
i ....... i " " " _ ......................
--"k
Objects
0 8 24 4 9 2 1 9 3 1 9 3 6 50
ONH B, fac(expl).
Figure 17b. Model ONH Scores Plot for Factor 1
50
0.050-
0.025
0
-0.025-
-0.050-
-0.075-
-0.100-
-0.125-
Lc,adings .................
I
X-variables
0 1300 1600 1900 2200 2500
ONH._B.B z factor.
Figure 17c. Model ONH Loadings Plot for Factor 2
1.5- Scor_ .......................
1.0-
0.5
.
-0.5 -
-1.0-
-1.5-
....... i ............................ t . . .
................................... ObieCts0' ......... 8 24' .........4 9 2' .........1 9 3' .........1 9 3' .........6 5'0
ONH B, fac(expl).
Figure 17d. Model ONH Scores Plot for Factor 2
51
0.10_Loading.s]
0.08-4
0.06 -
0.04-
0.02-
.
-0.02-
-0.04- v b es
6 13'00 16'00 1900 2200 2500
ONR Br factor.
Figure 17e. Model ONH Loadings Plot for Factor 3
2.0-
1.5-
1.0-
0.5-
0-
-0.5-
-1.0-
-1.5
-2.0-
Score_" ....................
.......>
...................................... , , ......... _ ....... ] .........
0 8 24 4 9 2 1 9 3 1 9 3 6 50
ONH B, fac(expl).
Figure 17f. Model ONH Scores Plot for Factor 3
52
0.20-
0.15-
0.10-
0.05 -
.
-0.05 -
-0.10-
-0.15-
I_x?adings
X-variables
1300 1600 1900 2200 2500
ONH B factor.
Figure 17g. Model ONH Loadings Plot for Factor 4
0.3-
0.2-
0.1
.
-0.1-
-0.2-
-0.3 -
SCO_S . . . . .............................
14('0%)_""""'":' i. i.'"'' ; i. "." i'.' i., ii: ?"" ..... i!'\ii" "_;'/q' l/_ i. i.
............ L ............ , .
................. objo ,.,......... i ......... l ........ i ........
........ 8 2_1 4 9 2 1 9 3 1 9__3__6 50
ONH B, fac (expl) .
Figure 17h. Model ONH Scores Plot for Factor 4
53
0.15-
0.10-
0.05-
.
-0.05-
-0.10-
-0.15-
Loadings
AElE
i ....... i
X-variables
0 1300 1600 1900 2200 2500
.O!111
Figure 17i. Model ONH Loadings Plot tbr Factor 5
0.4"
0.3-
0.2-
0.1-
0-
-0.1-
-0.2-
-0.3-
-0.4-
Seord ..................
i 1-
0 8244 9 2 1 9 3 1 9 3 6 50
t ONH B, fac_.
Figure 17j. Model ONH Scores Plot for Factor 5
54
6.5 Observations related to the reflected energy from the surfaces
Initial observations were made using a front surface mirror as the reference surface. We
soon learned that a mirror is not an appropriate reference for a steel surface. A series ofobservations were made illuminating the surface with a collimated beam and placing the receiving
fiber in the region that specular reflection would be expected. The patterns are shown in Figures18a through 18f Note that the mirror gave a specular reflection as expected. Reflection patterns
are shown for BaSO4 that is usually considered a diffuse standard for reflection. Reflection
patterns for D6AC steel at several angles are shown in Figures 18c through 18f. Specular
reflection accounts for a large percentage of the reflected energy.
The UV data reported for the study of contaminants was obtained at 450-45 o using the
reflection apparatus described above. The reminder of the data was obtained at approximately
0°-0 ° using a Guided Wave 19 fiber probe. This probe use nine fibers to illuminate the surface and
10 fibers to receive the reflected energy. The 10 fibers are configured to match the slit entranceto the detector.
When the collimated beam was being used to illuminate and observe the samples, very
rigid alignment was required. If the sample was not at the correct distance from the probe, largedifferences were observed for off axis illumination. Distance is not as critical for the 0-0 probe.
Recently we have found that alignment normal to the surface is very critical for this probe.
Specular reflection is believed to account for most of the signals observed during these
experiments. For example during the sequence studies, we noted that the energy from the sample
increased initially, then would decrease. This observation hypothesis would be consistent with an
hypothesis that the surface becomes glazed as a film forms on the surface or as reaction takes
place at the bottom of the pits on the surface.
Reflection spectroscopy has been used to measure surface roughness. It is possible that
some of the variance which is being observed is measuring surface roughness. A controlled study
where roughness is measured by an independent method should be beneficial for a complete
interpretation of the data.
55
Figure 18. Reflection patterns observed from various surfaces.
Figure 18a Mirror Surface at 45 °
li!i!iii!iiiiii!iii!iii!ii!:ii:i!i!ii
........ i
Figure 18b. Packed Barium sulfate at 45 °
Figure 18c. D6AC steel plate at 12 °
56
Figure 18d. D6AC steel plate at 45 °
Figure 18e. D6AC steel plate at 60°
Figure 18f. D6AC steel plate at 70 °
57
7.0 Conclusions
During the course of this research, we have worked with several types of data sets. The
first spectra were recorded using samples of compounds or contaminants similar to HD2 grease.
Since the spectrum of each compound could be obtained, it was possible to resolve the
compositions of mixtures of the compounds. As each set of environmental chamber experiments
were examined and the accumulated set of spectral observations grew, it became evident that the
technology we were using should also allow for the detection and identification of the species
resulting from oxidation and/or hydrolysis of D6AC steel. Much of the data that we have used
was obtained as sequential spectra recorded during extended periods in the environmental
chamber tests.
Since reproducibility is a necessity, the data were first examined as a series. For these
studies absorbance units were calculated by subtracting spectra of all later periods from the initial
spectrum in the series. Factor analysis was performed to determine the number of components
involved. Two to six were usually found. PCR and PLS2 were used in an attempt to determine if
the components always occurred in the same sequence. Models from each series were used with
the spectra from other series to determine if the order predicted were the same as the observed
order, i.e., one order can be used to predict the order of samples in another series.
In the analysis of these experiments, the time the predictions are not the same as the
observed values. For example, when using the data obtained at ~70 o to predict the data at 140°F
the predicted times are much greater than the observed times. This can be construed as
encouraging, since most chemical reactions proceed at a faster rate when the temperature israised. This observation is reinforced when we find that the 40°F samples have lower valeus than
predicted. We are encouraged that the same chemistry is involved. There is some evidence that
there is an induction period as a species begins to form, which is then followed by a fairly rapid
reaction.
Throughout all the aniyses of the D6AC surface spectra, we find reappearing spectral
features. We are encouraged to continue to work to identify the molecular species and the
mechanisms for the oxidations and other reactions that are occurring on the D6AC surface.
8.0 Optimization of the use of NIR for quality assurance
Building on the past experiences using NIR spectrophotometry coupled with chemometric
procedures investigations directed toward optimization of the use of NIR spectrophotometry
should be pursued. The following tasks are recommended.
1. Optimization of the techniques for examining surfaces. Primary emphasis for this task
would be measurement of corrosion products and organic contaminants on steel and aluminum
surfaces using a microprobe. The currently available 9/10 probe and Guided Wave Model 260
Spectrophotometer will be used to optimize experimental parameters and provide data for the
optimization of data reduction using multivariate algorithms.
58
2. Integration with materials analyses. Previous work with an ATK probe has indicated
that NIK spectrophotometry can be used to measure ratio of ingredients during processing and
that the data can be used to predict the physical properties of a material, for example bond
strength of the cured matrix, at the conclusion of the next processing step. The
spectrophotometer will be available for Task 1 to measure the properties of the surface. ATK
spectra would be taken of the matrix applied to the metal surface. The predicted strength of the
bond would be compared to the measured strength. Surface roughness measurements should also
be investigated.
3. Computerization of data reduction. New versions of the Guided Wave sottware and
Unscrambler II include microlanguages adaptable to "script" procedures. Matlab is also available
and includes script capability. Programs can be provided so that data is immediately available
during test thus providing an opportunity for immediate action.
4. Optimization of probe designs. Concurrent with Task 1 and 2 new probes would be
designed that optimize the data that is obtained. Hopefully, the newer Guided Wave INSIGHT
spectrophotometer will be available. This spectrophotometer has a much higher processing
capability and more importantly, a noise level at 0 AU of less than 20 nicroabsorbance units.
Recent tests have indicated that the average of 32 spectra has a noise level of 4 microAU. These
32 spectra can be taken quicker than 1 spectrum with the Model 260, thus a further indication of
how technology is improving our ability to monitor processes.
9.0 Acknowledgements
A number of students and staff have assisted in carrying out this research project. Ms.
Yadilett Garlington, Ms. Bianca Brindley, Mr. Brian Benson, Mr. Markus Heilbrotter, and
Morgan Wang, all contributed to some part of this activity. Each is to be commended for their
assistance.
10.0 References
1. Martens, H. and T. Naes, Multivariate Calibration, Wiley, New York, 1989.
2. Handbook of Military Infra-red Technology, Office of Naval Research 1966.
3. Weyer, L. G. "Near-Infrared Spectroscopy of Organic Substances," Applied Spectroscopy
Review 21 (1985), 1-43.
4. Massart, D.L., et. al. Chemometrics: a textbook, Elsevier, New York, 1988.
5. Malinowski, E. R., Factor Analysis in Chemistry, Wiley, New York, 1991.
6. DeNoyer, Lin and Dodd, Jack, "Square Tools," Spectrum Squares Associates, Ithaca, N.Y.
1992.
59
7.Hauffe, K., Oxidation of Metals, Plenum Press, New York, 1965.
8. Hecht, H.G., Infrared-Spectroscopy,
9. Greenler, R.G., lout" Chem Phys 44 (1966) pp 310 - 315.
10. Greenler, 1LG., Jour Chem Plays 50 (1969) pp 1963 - 1968.
11. Greenler, R.G., Surface Scien 69 (1977) pp 647-652.
12. Allara, D.L., et. al., Macromolecules 11 (1978)pp 1215 - 1220.
13. Martens, H. and B. Alsberg, Analytical Applications of Spectrsocopy II, Roy Soc. of Chem,
Cambridge, 1991, pp 221 - 239.
14. Naes, T. and T. Isaksson, Computer-Enhanced Analytical Spectroscopy, vol. 3, Plenum Press,
New York, 1992, pp 69 - 94.
15. Ares, J., Anal. Chem. Acta, 268 (1992) pp 135 - 144.
60
APPENDIX. Study ofmixtures offourhydroxyl containingcompounds
As indicated previously analyses of spectra of mixtures of known compounds is
straightforward. Several methods are suitable for determining the calibration matrix. Since we
were interested in the wavelengths at which the OH group absorbs, a set of 8 mixtures was
prepared using methanol, ethanol, isopropanol, and glycerol. (Table 6 shows the mole percentage
of each alcohol sample) The NIR spectra were recorded for the same wavelength region as used
for the other studies in this report (Table 7 shows some groups and their wavelengths). The raw
data is presented in Figure 19a through 191. A calibration model was set up using PLS2 from
UNSCRAMBLER with this data and the spectra of the pure alcohols. The compositions were
expressed in mole fractions of the mixture. This reduces the degrees of freedom by one since the
mixtures must add to 1.0. Figures 20a through 20k show the predicted results vs the known
compositions.
Even though the loadings can not be considered "real" spectra it is informative to compare
the loadings to the spectra of the pure materials. Factor 1 and Factor 2 are compared to methanol
and glycerol in Figure 21a and 21b. Figure 22 shows that glycerol is explained largely by the
variance from Factor 1. Looking at the chemical formula for glycerol, CI-I:OH=CHOH-CHzOH,
it is clear that it contains all of the chemical vibrations that occur due to C-C, C-H, or O-H
vibrations that may occur in the selected alcohols. Because the vibrations in glycerol comprise the
variations common to all of the pure alcohols in the mixtures, it is evident that Factor 1 would be
most similar to the spectrum of glycerol. Figure 23a through Figure 23h are the loadings and
Figure 23a through Figure 2h are the scores.
All the prior work presented in this report, except for the contaminant study, was based on
unknown chemistry. Hence, we felt that an informative experiment based on known mixtures
would assist in building confidence in Unscrambler II as a calibration tool. Consequently, the
following study was performed to verify that the components of a known mixture can be resolved.
61
TABLE VI. MOLE PERCENTAGE OF RESPECTIVE ALCOHOL
OBJECTS
SOL 01 1
SOL 02 2
Methanol Ethanol
1 0 0
0 1 0
1
Propanol
SOL 03 3 0 0
SOL 04 4 0 0 0
SOL 05 5 0.59 0.41 0
0.66 0 0.35
0 0.57 0.43
0.21
SOL 06 6
SOL 07 7
SOL 08 8 0.4 0.28
SOL 09 9 0.45 0.31 0.24
SOL 10 10 0.25 0.35 0.4
Glycerol
0
0
0
1
0
0
0
0.11
0
SOL 11 11 0.53 0.19 0.28 0
SOL 12 12 0.55 0.25 0.1 0.11
TABLE VII. Some Groups and Their Wavelengths
Group Overtone Wavelength Intensity
C-H 1 1700 nm strong
2 1100 nm medium
O-H 1 1400 nm strong
C-C 3 1750 nm very weak
4 1400 nm not detectable 1_
C-H 1 1600 - 1800 nm N/A
2 1100 - 1250 nm
1_ Set Combinations 2000 - 2400 nm
2 "_ Set Combinations 1300 - 1450 nm weaker
O-H 1 1400 - 1416 nm strong
2 1000 nm weaker
Combination bands 2000 nm weaker 3
_Wheeler, Owen H., "Near Infrared Spectra of Organic Compounds," Chemical Review, 59,
629-666 (1959).
'Approximate theoretical wavelengths of overtones.
3Weyer, L.G., 'rNear-Infrared Spectroscopy of Organic Substances," Applied Spectroscopy
Reviews, 21(1&2), 1-43(1985).
62
.4 -
1.2-
1.0-
0.8-
0.6-
0.4-fi
!<SOL01>Figure 19a. Raw data of SOL01
1.25-
1.00-
0.75-
0.50-
0.25 -
i ..... ii i !
....... : ....... : ....... - ............... :
h<SOL02>
i a'oa 1grin 1oho 7.9.'00 _4aa
Figure 19b. Raw data for SOL02
ii,1 i i i iii i,o iii iri .... iiiii i iiiiii i0.9- i
080.7-
0.6-
0.5-
0.4-1] I"_'f}(} I gl'l(l I qf}(} 9.9_flD _fl(l
<SOL03>
Figure 19c. Raw data of SOL03
63
1.4-
1.2-
1.0-
0.8-
0.6-
0.4-
....... i ....... i GLYCEROL " i ....... i ....... i
Figure 19d. Raw data of SOL04
1.4_
1.2-
1.0"
0.8_
0.6-
0.4-_
....... i ....... METHANOL 10 mL_ ETHANOL 10 IUL i ....... i
i
i
I
I6
i<SOL05 >
Figure 19e. Raw data for SOL05
1.4
1.2
1.0-
0.8
0.6-
0.4-_
<SOLO riO>
: 191_OE10_PROPANOL10mL " " " i ....... i
iii .......................................
Iq'f)fl IgfW') 10'flO 9.9.'1"10 "_'O0
Figure 19f. Raw data of SOL06
64
1.2-
0.9-
0.6-
0,3-
<SOLO_>
....... i .
i.......i....iiFigu_ 19g. Raw dataofSOL07
1.4-
1.2-
1.0-
0.8-
0.6-
0.4-6
<SOL08>
" ' • METHANOL 1-0mL; ETHANOL 10mL,-PROPANOL 10mL; GLYCOL 5 mL .... :
1"_ I gt_ 1_tm _.:_iao 9gaa
Figure 19h. Raw data for SOL08
1.4
1.2-
1.0-
0.8-
0.6-
0.4-
<SOLQ_>
....... METHANOL 10 _ _OL 1"0ml_ PROPANOL10 m.L ....... i
L :: •
I q'I3{) IgfIO 1 _ _)]t"_ ).._'f)l')
Figure 19i. Raw data of SOL09
65
1.4-
1.2-
1.0-
0.8-
0.6-
0.4-
<SOLIO>
........ :METHAJqOL 5 ml_ ETHANol.' I0 ml_ PROPANOL iSn_ ....... :
.......... : ....... : ....... :
Figure 19j. Raw data of SOL10
1.4-
1.2-
1.0-
0.8
0.6-
0.4-
i ....... : " METHANOL I.'0ml_ ETHANOL 5mL, PRoPA_qOE 10niL ...... :i
6<SOL1 i>
1"_'fg'l 1 fi'fl(I 10'(_ _9'fl(I 9__'{'1(}
Figure 19k. Raw data for SOLll
1.41.6" .... ME_OL _OL 10 .m_.- PR01ANOL 15 .i_.., GLY!OL. 5.3 mL.
1.2
1.0 °
0.8-
: ! ! : !,6 .........................................
6 1_,'f}fl Igig) 160N 99fin 9__NN<SOL12>
Figure 191. Raw data of SOL12
66
0.,38- P_,
0.37-
0.36 -
0.35-
0.34 -
0.33 -
TEST n
pdicted (with 1-factors) ...... ; ....... _ ........tf'L
: lu+ 9 11 5i 6 i :
'"l
: -
: Z
: 12:: ....... ! ....... !- ; :
i....... : ....... : ............... : ....... -
....... ; ....... ; ....... _ ....... - ....
h 0'2 a',l O'_ D'8 +"nY_,u.m I,'_- kTR._N]_I_T-_T.
Figure 20a. Predicted vs. known compositions of Methanol using 1 factor
1.0- Pn
0.8-
0.6-
0.4-
0.2-
0-
-0.2-
/iTEST B.
:,._,e,, _,,t. ,...oc,_rs)a:.. A/...: h-, ¢..^ x ...... : ....... : ....... : ....... :
i i 12 i i i
i /
........ : ............................. m sr_u"'ea-u -_
0'2 0'4 fl'_ 018Y-va r : METHANOL
Figure 20b. Predicted vs. known compositions of Methanol using 2 factors
110
1.0-
0.8-
0.6-
0.4-
0.2-
0-
-0.2-
!TEST
Pu
3
B°
_dicted (with 3.facto.m) ....................... :
....... i ....... i .................. :....... .......
....... : ....... ! ....... i ....... ! .... :•Measured
Y-vat : METe''_OL fl'4 fill n*R
Figure 20c. Predicted vs. known compositions of Methanol using 3 factors
1:o
67
1.2-
1.0-
0.8-
0.6-
0.4-
0.2-
O-
TEST B.
Pr dictexl (witk 4 fa_o.rs) ....... : ....... : ....... : ....... :
......... _'_,, ......... i ....... i
..... _. ........... i ....... i ....... _
fl'2 fl'g fl'fi a'_ l'nY-vart METHANOL
Figure 20d. Predicted vs. known compositions of Methanol using 4 factors
1. 0 -
0.8-
0.6-
0.4-
0.2
O-
P rdict with5 acto. ............ii.......i.......i.......i.......i ......i....... : ....... 8.... 115i ..... i ....... i
;, Measured
I_i_ f114 fl'fi 1"118 1_fl_RR_ 18. Y-vat t MET_OT,
Figure 20e. Predicted vs. known compositions of Methanol using 5 factors
1.0-
0.8-
0.6-
0.4-
0.2-
0-
-0.2-
TEST B.
....... : ....... : ....... : ....... : ..... M_su/ed
Y-var 1
Figure 20f. Predicted vs. known compositions of Methanol using 6 factors
68
1.0-Pn
0.8-
0.6-
0.4-
0.2"
O- q tVteasmextN N'2 fl'4 N'6 N'fl I'0
TEST B. Y-var t METHANOL
Figure 20g. Predicted vs. known compositions of Methanol using 7 factors
1.0- Pn,,dicted (with-8 factors) .......................... • . - - _...1
i i ! i i __:0.8 i : i i i i
0.6-
0.4- i ! ! i i i0.2-
O- i _d
fi n'_. n'4 o16 n'_ i'.oTEST B. Y-vat t METHANOL
Figure 2Oh. Predicted vs. known compositions of Methanol using 8 factors
69
1.0- Pr
0.8-
0.6-
0.4-
0.2-
0-
NTEST la.
MeasuredN'2 N'4 0'6 O'fl
Y-vart ETHANOL
Figure 20i. Predicted vs. known compositions of Ethanol using 8 _ctors
1'N
1.0- Pr
0.8-
0.6-
0.4-
0.2-
0-
0TEST B.
l_easmvxlfl 2 fl 4 0 fi ilia
Y-var -- PROPANOL
Figure 20j. Predicted vs. known compositions of Propanol using 8 factors
lln
1.0- Pr
0.8
0.6-
0.4-
0.2-
0- 1
0TEST B.
d
¥-var - GL_EROL 0'4 0'6 N'fi
Figure 20k. Predicted vs. known compositions of Glycerol using 8 factors
lZn
7O
-0.015-
-0.020-
-0.025 -
-0.030-
-0.035-
-0.040-
-0.045-
..................................., !i.B r
', :?;1 ;" t
.... 2! • " : ....... :' ........... • - . : .......... ,::. -: . •
i ." • ."'. i '_ " ." . . -:1 • ,- , • • . /; - ,. . ', ..: • \ : '. . . • ,. . . .
ii ' ' '.,' '_ - " • ! •
'. ' ' " ; " /",..: • k..,'"--"
•..,.:" _.:' ":"... :..._ _ : : : :"! • '.i " :
ki : :........ : ....!. ! ........ 1"= Sdali_d.Methaiaol $1_¢fi.ui n ...... :
: " 2= Factor 1:
t :
........................................ !
6 131_3.0 16(JO.O 19d0.0 221J0.0 25(J0.1_
<13-0.035> <Pax Fac 001>i
Figure 21a. Methanol spectrum compared to Factor 1
0.18-
0.15-
0.12-
0.09-
0.06 -
0.03 -
O.
-0.03-
-0.06-
,.- ...... : ....... : ....... : ...............i . /N, . "
• /,/\ . 1 = Scaled Methanol Spectrum .
...... !ii\ 2: i.or " ...... .......
" • I:i, . : :" :',';" ' • "i" " : .......................
I ,7:::"h /i _: ;.- ::' :: : . : i_ i--\j_- "<, '...."_ , ,. -"'-., . ., _i,,.,, : .,.,' "...,., :, :'.,_,.: -, . :'., : ,,
...... "; ......... : . ."':":". :.1. : ....... : .... _'I • \. :
I _ ' l ' ' I ' ' _ I ' ' I ' ' I
0 1300.0 1600.0 1900.0 2200.0 2500.13
<115-0.1> <Pax Fac 002>m
Figure 2lb. Methanol spectrum compared to Factor 2
71
-0.015-
-0.020-
-0.025 -
-0.030-
-0.035-
-0.040-
-0.045-
r ,, .... : i "
J .... i.:........................ /".:. :..• . •
/.
p', , _ °/°_,.-
, '- - 4: • -:--..,'...... ; ............................•-,,,:.:
•t : ", : ',
....... : ".."! i ' • " ! • ' • 1 = Scaled Gly_rol Sl_ctrutn ": : 2=Facto/1
i :.......... '," i .............................
m
"!
'0 13'00 1600 19'00 22'00 25'00
<15-0.035> <Pax Fac 001>
Figure 21c. Glycerol spectrum compared to Factor 1
0.175-
0.150-
O.125-
0.I00-
0.075-
0.050-
0.025-
0-
-0.025-
<4/5-0.1>
......... .... ....... : ..............
%!
! ..,,' ',,,,' : :: ;_,.." ,, ,..: ,,,: '.. .
'," , , I,' '_ ,
6 13'00 16'00 1900 22'00 25'oo<Pax Fac 002>
Figure 21d. Glycerol spectrum compared to Factor2
72
100-
80-
60-
40-
20-
_
Fac 000
% Y:va.rian .ce expl.. .............................
• . °.-" • t I ....
.." . / . . .
....... b. o ......... . • . • ...................
• ." • - .- • r ....
................. ,':......... iii : :i . . . ...".... :.:. ." , ....
: " " i t , i i . . .
! : . ; . . " / _ .....
; . .." . . / .....
/ • . . t .....
: . . . i I .....
,' . . . I .....
.' ! ....
":' .... ':'" " " i .... i ",/" " " i .... i .... i .... i .... i,' ' . • I .....
/ / • ;" . . / .....
- - OL ;'Factors
I ......... I ......... I ......... I ......... I ......... I ......... I ......... I ......... I
Fac 001 Fac 002 Fac 003 Fac 004 Fac 005 Fac 006 Fac 007 Fac 00
TEST B, variable.
Figure 22. Percent Variance Explained
73
-0.015-
-0.020 -
-0.025-
-0.030-
-0.035-
-0.040-
-0.045-
te_t b.
Ix
.................................... X-va dablesh 1_{_1 rl 16d_3 CI lOCN1 n _?_ n _{_ f
factor.
Figure 23a. Factor 1 of Hydroxyl Mixture Set
Figure 23b. Factor 2 of Hydroxyl Mixture Set
0.12-_
0.10-
0.08
0.06-
0.04-
0.02-
0-
-0.02-
-0.04 -
test b.
,adings •
! i i iii!! il !ii•factor.
Figure 23c. Factor 3 of Hydroxyl Mixture Set
74
0.09-0.06-0.03-
0-43.03-43.06 -
43.09-
-0.12-
43.15-
tagt b.
Ix adings. • .^ • : ....... : ....... : ................
Z
• ; . . .
........ : ....... : ....... : ....... : .... X-variabies
0/acre r . I"_I_lII I_ N 10t'_tN 79f_ N 9_tiNt_
Figure 23d. Factor 4 of Hydroxyl Mixture Set
0.18
0.15-
0.12-
0.09-
0.06-
0.03 -
0-
43.03 -
-0.06-
test b.
_, a,,,,,# ..... : ....... : ....... : ....... : ...... :
i ....... : ....... : ....... _ ....... ; ....... :
X vanablesii i ii iil i ii: ii iii i i_. "!6 1"_rinn 1r_n n _on n "_rin n "_ rfactor.
Figure 23e. Factor 5 of Hydroxyl Mixture Set
0.06 -
0.04-
0.02-
0-
-0.02-
-0.04-
43.06-
43.08 -
43.10-
Lcadings " I " i .............. i ....... i ....... i
............... : ....... : ....... : .... X-variables
test b. 0/actor. l'_tiNN I(-_'_N IO_N "_TfiON 7_;t_f
Figure 23f. Factor 6 of Hydroxyl Mixture Set
75
0.075-
0.050-0.025-
0-
-0.025--0.050--0.075-
-0.100-
t_t b.
Lc adings ..... : ....... : ....... : ....... : ....... :
_ J.... i....... i ....... i ....... ',....... i....... : ....... : ....... : ....... : .... X-variabies
_aetor. 1"_finn 1_ n 1ofin n _._ n 9_fin e
Figure 23g. Factor 7 of Hydroxyl Mixture Set
0.100-
0.075 -
0.050 -
0.025 -
0-
-0.025 -
-0.050 -
-0.075-
-0.100-
test b.
l.,cadings- _.... : ....... : ....... i ....... :: ...... i.... i ....... i ....... i ....... i ....... i
....... : ....... : ....... : ....... : .... X-variables
fOaCtO r . I afiO rl l_n n 10/_3 0 99(_113 ?qfiO f
Figure 23h. Factor 8 of Hydroxyl Mixture Set
76
5.0-
2.5-
0-
-2.5 -
-5.0-
-7.5 -
-10.0-
-12.5-
S( )rP.,S ..... - ...... - ...... " ...... " ...... "- .....
...... i ...... : ...... ! ...... i ...... i .... obie_.qf3l .fl_ ROT fldl. RC3T _ _NT .fiR .qG31.113 Rt"_T 1 :
l
Figure 24a. Scores for Factor 1 of Hydroxyl Mixture Set
1.5-
1.0-
0.5
0-
-0.5 -_
-1.0-
-1.5-
*I'IR_"P
_2.%) _ _ _ _ _ i
...... ! ...... ! ...... ! ...... i ...... i .... ."
13 NOT .tq9 ROT .fkl NOT/_6 ROT .OR ROT .10 .qtq1.1
Figure 24b. Scores for Factor 2 of Hydroxyl Mixture _et
2.5-
2.0-
1.5-
1.0-
0.5
0-
-0.5--
-1.0-
-1.5-
T1RS_ lB.
SC :)I'P.,S ..... : ...... : ...... : ...... : ...... : ...... :
• ObleCts0 ROT .f)2 ROT .Od .qCIT f16 .'qOf .fir ROT .1 f} .qf_T .1
fa,- I _xt)1% .
Figure 24c. ScoresforFactor3 of Hydroxyl Mixture Set
77
0.6"
0.4-
0.2-
0-
-0.2
-0.4-
-0.6-
Sc _re8 ............ : ...... : ..... : ..... : ..... :
............ i ...... _...... i ...... i ...... i
i i i_ Iii I __ I I I I I I ;14(0_ i i i I I i I Iii I I ii II: I I II----
Figure 24d. Scores for Factor 4 of Hydroxyl Mixture Set
0.4-
0.3-
0.2-
0.1-
0-
-0.1 -
-0.2-
-0.3 -
-0.4 -
TEST B.
sc,_res.5(Q%).' i ...... i ...... ::" " ^ " " i ...... i ...... ::
I ...... : ...... : ...... : ...... : ...... : .... Obje_fl ._ni.o_. _ni.rut ._ni _ _qni J),q .qni .1N _qni .1 ",
fac f exD1 _.Figure 24e. Scores for Factor 5 of Hydroxyl Mixture Set
0.2-
0.1-
0-
-0.1 -
-0.2-
-0.3 -
-0.4 -
-0.5 -
-0.6 -
TEST B. fac
Sc _ r_ ..... : ...... : ...... : ...... : ...... : ...... :
ii!iililil i...... i ...... i ...... i ............. i .... .:
_l CI4 _NT fl_ _l fir _l 1fl
Figure 24f. Scores for Factor 6 of Hydroxyl Mixture Set
78
0.2- Sc _res ..... : ...... : ...... : ...... : ...... : ...... :
...... i.... i...... i...... i...... i
iiiiiiiiiiiiiiiill ....!....ob,e ._NT N_ RNT 1"_ ._NT N_ RNT .NR _NT _1N _t")1.1
"!_ R_ f._ IAT_L'_ .
Figure 24g. Scores for Factor 7 of Hydroxyl Mixture Set
0.10-
0.05-
0-
-0.05 -
-0.10-
-0.15-
q't_.._flll I_ _
Sc ) re.,s ..... : ...... : ...... : ...... : ...... : ...... i
!l ° . ° "" /......i............ i......i......i......i.................... : .................. Obieets
_ I _art_l 'l .
Figure 24h. Scores for Factor 8 of Hydroxyl Mixture Set
79