🚀 go-pugleaf

Just before Christmas I posted the following message:

      ----------------------------------------------------------------------
      I am assisting a colleague in the supervision of a senior student
project.
      
      A local dairy company uses Dickey-john Near Infra Red analyzers to
      estimate %fat, %moisture, %protein, etc in dairy products. The
      analyzer uses a multiple regression equation with predictors being
      the logs of reflectance values at several frequencies. From time to
      time the analyzers are calibrated against standard chemical methods.
      
      In between calibrations minor adjustments are made on the basis of
      the results for a few items. The student's task is to review the
      procedure being used for these adjustments, and for the calibrations
      themselves.
      
      We would be interested in any advice, or pointers to appropriate
      literature, that someone who has worked in this area previously
      might be able to give.
      -------------------------------------------------------

I didn't expect much of a response in view of the time, but I was anxious
to get started on the hunt for references. I was pleasantly surprised by the
useful responses that I received.

Joseph Lebouille [lebouille@voeding.tno.nl] writes :

       Two books which I strongly can recommend to you are:
       Near Infrared Technology in the Agricultural and Food Industries
       by P.Williams and Karl Norris. Published by: American Ass of
       Cereal Chemists Inc.1987

       The more sophisticated way of calibrating NIR's nowadays is 
       by PCR or PLS-1. The book I can recommend is:
       Multivariate calibration by Harald Martens en Tormod Ns
       Published by John Wiley & Sons, 1989 (ISBN 0 471 90979 3).


Jay Goodman  [goodman_jh@alcoa.com] writes

      You may be interested in:

       Application of Near Infrared Reflectance Spectroscopy to the
       Compositional Analysis of Biscuits and Biscuit Doughs

      Osborne, B. G., Fearn, T., Miller A. R. and Douglas, S.
      J. Sci. Food Agric. 1984, 35, 99-105      and,

       Multivariate Calibration
      P. J. Brown,      JRSS B, 1982, 44, 3, 287-321


Aloke Phatak          [phatak@watserv1.uwaterloo.ca]  writes

      Try the following book:

      Martens, H. and Naes, T. (1989). Multivariate Calibration. 
      John Wiley and Sons:Chichester.

      Martens and Naes discuss calibration of NIR instruments using 
      full spectrum methods like principal component regression (PCR)
      and partial least squares (PLS).  They claim that full spectrum
      methods yield better predictions than methods based on selecting
      particular wavelengths.  Nice discussion of outlier detection, etc.


Peter Wollan       [wollan@mayo.edu]        writes

      Having just run into a horrific instance of ignorance about basic
      quality control notions, I'd like to encourage your student to read
      Deming--any of his stuff, really, but in particular he discusses the
      reasons not to re-adjust based on individual or too-frequent values. 
      It's in each of his books--he's a fairly repetitive writer--"Out of
      the Crisis" is most recent.


I have had a bit of a read of the 1982 JRSSB paper of P.J. Brown and it is
quite interesting to see that inverse regression for each of the
compositional variable separately has its defenders, which is just as well
because this seems to be the standard industrial practice. There is an
extensive discussion at the end of this paper which is helpful in
generating further references. D.R. Cox, in his contribution raises a point
which is very relevant to our present project: 'A crucial aspect in
applications is often the stability of the calibration curve in time: how
often is recalibration needed, and how can checks of stability be
incorporated into the routine use of the procedure?' In his response to this
question, Brown cites what looks like a very relevant paper -
Daniel, C. (1975) 'Calibration designs for machines with carry over and
drift' J. Qual. Tech, 7, 103-108.

I also checked Christine Osborne's useful survey paper in International
Statistical Review, v59, n3, 309-336 (1991). This has lead me on to further
references including

Brown & Sundberg (1987)  JRSSB 49, 46-57.
T. Fearn(1983) Misuses of ridge regression in the calibration of a near
infrared instrument. Applied Statistics, 32, 73-79.
T. Naes (1986) Multivariate calibration using covariance adjustment.
Biometrical J. 28, 99-107
T. Naes, C. Irgens, and H. Martens(1986) Comparison of linear statistical
methods for calibration of NIR instruments. Applied Statistics 35, 195-206.

Clearly there is a lot of work to be done tracking down and digesting all
this literature, nevertheless we would still be grateful for further sources
and comments on any of the references mentioned.
-- 
Murray A. Jorgensen [ maj@waikato.ac.nz ]    University of Waikato
Department of Mathematics and Statistics     Hamilton, New Zealand      
__________________________________________________________________
'Tis the song of the Jubjub! the proof is complete,
if only I've stated it thrice.'