Howard_Mark_chemometrics applied to NIR spectroscopy

chemometrics

1. “Old Timer’s” Night
Howard Mark
IDRC 2014
Chambersburg, PA

2. Highlights
•3/12/1942 – Born
•3/1972 – PhD, New York University
•9/1976 – Joined Technicon Industrial Systems – Studied Statistics
•1985 - Published Mahalanobis Distances (Anal. Chem. 57, 1149 (1985))
•1986 – Published Normalized Distances ( Anal. Chem., 58(2), 379- 384(1986))
•1986 to Present - Spectroscopy Columns
•1989 – Accuracy of individual analytical methods (Anal. Chem.; 61(50,
pp. 398-403 (1989))
•1991 - Technicon closed down – began consulting, developing products
•2002 - Noise in calibration (Appl. Spect ., 56(5), p.633-639 (2002))
•2010 - Inherent non-linearity in NIR (Appl. Spect ., 54(9), p.995 (2010))

3. The face that got him stuffed into lockers in High School

4. Highlights
•3/12/1942 – Born
•3/1972 – PhD, New York University
•9/1976 – Joined Technicon Industrial Systems – Studied Statistics
•1985 - Published Mahalanobis Distances (Anal. Chem. 57, 1149 (1985))
•1986 – Published Normalized Distances ( Anal. Chem., 58(2), 379- 384(1986))
•1986 to Present - Spectroscopy Columns
•1989 – Accuracy of individual analytical methods (Anal. Chem.; 61(50,
pp. 398-403 (1989))
•1991 - Technicon closed down – began consulting, developing products
•2002 - Noise in calibration (Appl. Spect ., 56(5), pp.633-639 (2002))
•2010 - Inherent non-linearity in NIR (Appl. Spect ., 54(9), p.995 (2010))

5. Highlights
Post-Technicon
•Consulting for Pharmaceutical, Chemical, Electronic, Agricultural, other
Industries
•Product development – Software
–All-Possible-Combinations wavelength search with MLR (quantitative)
–All-Possible-Combinations wavelength search with Mahal. Dist (qualitative)



–Conformity Index




–Linearity test

6. Highlights
Post-Technicon
•Consulting for Pharmaceutical, Chemical, Electronic, Agricultural, other
Industries


•Product development - hardware
–High-pressure cells
–Fiber-optic Adaptor
–Grey reflectance material

7. High-Pressure Cells

8. Linearity in Calibration

9. Non-Linearity: Wt % to Vol %
Toluene
Dichloromethane
N-heptane

10. What we discovered
1)Weight % and Volume % are nonlinearly related
2)Spectroscopy is inherently sensitive to VOLUME percent
3)Therefore, spectroscopic measurements are inherently
non-linearly related to weight percent

11. Implications & Ramifications
1)This finding explains many of the “mysteries” we’ve been
fighting all these years:
A)Need for too many factors
B)Error measures do not go to the noise floor when # factors = # variables
C)Reason why spectral analysis considered good only for “dilute solutions”
D)Reason why calibrations have not been extrapolatable
E)Reason calibration transfer is so difficult.
2)The CLS algorithm is the nearest thing we have to an
“absolute” analytical method. therefore results are obtained
without the need for conventional calibration.
A)Reference Lab error is not an issue
B)CLS results are obtained with zero PLS or PCR factors
C)“Overfitting” is not possible
3)You can multiply volume percent by sample properties (e.g.,
density, MW) so that conventional algorithms can use other
measures of concentration, that contain a volume in the
denominator (e.g., weight / volume, molarity), which will then
be linear with respect to the spectroscopy.
4)Sum of CLS calculated concentration = inverse of Partial Molal
Volume