Final Report November 1, 1992 · Final Report submitted to NATIONAL AERONAUTICS AND SPACE...

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 ............................................


Loadings Plot for Factor 7 .........................................

Scores Plot for Factor 7 ...........................................


Loadings Plot for Factor 8 .........................................Scores Plot for Factor 8 ...........................................


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 ONtI Scores Plot for Factor 2 ..............................

Model ONH Loadings Plot for Factor 3 ............................Model ONH Scores Plot for Factor 3 ..............................





Mirror Surface 450 .................................................

Packed Barium sulfate at 45 o ......................................

D6AC steel plate at 12o ............................................

D6AC steel plate at 45 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










of Methanol using 1 factor .......

of Methanol using 2 factors ......




of Methanol using 6 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 6 of Hydroxyl 1W_txture Set


Factor 8 of Hydroxyl Mixture Set ..................................

Scores for Factor 1 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


September 16, 1992

September 16 - 17, 1992


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


September 28, 1992

September 28 - 29, 1992

" September 29 - August 12,1992


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

Final Report November 1, 1992 · Final Report submitted to NATIONAL AERONAUTICS AND SPACE...

Documents

Transcript of Final Report November 1, 1992 · Final Report submitted to NATIONAL AERONAUTICS AND SPACE...