Method for correcting spectral data for data due to the spectral measurement process itself and estimating unknown property and/or composition data of a sample using such method

• US 5,121,337 A
• Filed: 10/15/1990
• Issued: 06/09/1992
• Est. Priority Date: 10/15/1990
• Status: Expired due to Term

• Current Assignee: Exxon Research and Engineering Company (Exxon Mobil Corporation)
• Original Assignee: Exxon Research and Engineering Company (Exxon Mobil Corporation)
• Inventors: Brown, James M.
• Primary Examiner(s): CANGIALOSI, SALVATORE A
• Application Number:
  US07/597,910
• Time in Patent Office: 603 Days
• Field of Search: 364/498, 364/571.01, 364/571.02, 364/571.03, 364/571.04, 364/571.05, 364/571.06, 364/571.07, 364/571.08, 364/574, 364/578
• US Class Current: 702/28
• CPC Class Codes: G01N 21/274 Calibration, base line adjustment, drift correction
• CPC Class Codes: G01N 21/3504 for analysing gases, e.g. multi-gas analysis
• CPC Class Codes: G01N 21/3577 for analysing liquids, e.g. polluted water

Abstract

For correcting measured spectral data of n samples for data due to the measurement process itself, e.g. due to spectral baseline variations and/or water vapor and carbon dioxide present in the atmosphere of the spectrometer used to make the spectral measurements, the spectral being quantified at f discrete frequencies to produce a matrix X (of dimension f by n) of calibration data, matrix X is orthogonalized with respect to a correction matrix U_m of dimension f by m comprising m quantified correction spectra, at the discrete frequencies f, which simulate data arising from the measurement process itself. The correction method is preferably included in a method of estimating unknown property and/or composition data of a sample under consideration, in which the n samples are calibration samples and a predictive model is developed interrelating known property and composition data of the calibration samples to their spectral data corrected for the data due to the measurement process itself. Then, the unknown property and/or composition data of the example under consideration is estimated from the predictive model on the basis of its measured spectrum.

Claims

1. A method of estimating unknown property and/or composition data of a sample, comprising:

(i) collecting respective spectra of n calibration samples, the spectra being quantified at f discrete frequencies (or wavelengths) and forming a matrix X of dimension f by n;

(ii) producing a correction matrix U_m of dimension f by m comprising m digitised correction spectra at said discrete frequencies f, said correction spectra simulating data arising from the measurement process itself;

(iii) orthogonalising X with respect to U_m to produce a corrected spectra matrix X_c whose spectra are each orthogonal to all the spectra in U_m ;

(iv) collecting c property data, composition data, or property and, composition data for each of the n calibration samples to form a matrix Y of dimension n by c (c>

1);

(v) determining a predictive model correlating the elements of matrix Y to those of matrix X_c ;

(vi) measuring the spectrum of the sample under consideration at said f discrete frequencies to form a matrix of dimension f by 1; and

(vii) estimating the unknown property and/or composition data of the sample under consideration from its measured spectrum using the predictive model.

2. A method as claimed in claim 1, wherein the predictive model is determined in step (v) using a mathematical technique to solve the equation Y=X_c^t P+E, where X_c^t is the transpose of X_c, P is a prediction matrix of dimension f by c, and E is a matrix of residual errors from the model and is of dimension n by c and wherein, for determining the vector y_u of dimension 1 by c containing the estimates of the c property and/or composition data for the sample under consideration, the spectrum x_u of the sample under consideration, x_u being of dimension f by 1, is measured and y_u is determined from the relationship y_u =x_u^t P, x_u^t being the transpose of spectrum vector x_u.

3. A method as claimed in claim 2 for which f is less than or equal to n, wherein said mathematical technique is linear least squares regression.

4. A method as claimed in claim 2, wherein said mathematical technique is principal components regression.

5. A method as claimed in claim 2, wherein said mathematical technique in partial least squares regression.

6. A method as claimed in claim 1, wherein the spectra collected in step (i) and the spectrum measured in step (vi) are all absorption spectra.

7. A method as claimed in claim 1, wherein the collected spectra of the calibration samples and the measured spectrum of the sample under consideration are in the near-infrared range.

8. A method as claimed in claim 1, wherein the collected spectra of the calibration samples and the measured spectrum of the sample under consideration are in the mid-infrared range.

9. A method as claimed in claim 1, wherein the m spectra forming the columns of matrix U_m are all mutually orthogonal.

10. A method as claimed in claim 1, wherein step (ii) comprises:

(iia) modelling baseline variations by a set of orthonormal frequency (or wavelength) dependent polynomials which form the columns of a matrix U_p of dimension f by p where p is the order of the polynomials and the columns of U_p are orthonormal polynomials;

(iib) supplying at least one ex-sample spectrum which is representative of an anticipated spectral interference due to an ex-sample chemical compound, this ex-sample spectrum or these ex-sample spectra forming the column(s) of a matrix X_s of dimension f by s where s (>

1) is the number of such ex-sample spectra and is greater than or equal to the number of such ex-sample spectral interferences s'"'"';

(iic) orthogonalising the column(s) of X_s with respect to U_p to form a new matrix X_s '"'"';

(iid) orthogonalising the column(s) of X_s '"'"' among themselves to form a new matrix U_s ;

(iie) combining matrices U_p and U_s to form a matrix U_m whose columns are the columns of U_p and U_s arranged side-by-side.

11. A method according to claim 1, wherein, for determining any significant change in the measurement process data between the times steps (i) and (vi) were performed, the following steps are carried out:

(a) a matrix V_m of dimension n by m is formed as the dot product X^t U_m, where X^t is the transpose of matrix X;

(b) the corrected data matrix X_c is formed and its singular value decomposition computed as UΣ

V^t, where U is a matrix of dimension f by n and contains the principal component spectra as columns, Σ

is a diagonal matrix of dimension n by n and contains the singular values, and V is a matrix of dimension n by n and contains the principal component scores, V^t being the transpose of V;

(c) a regression of the form V_m =VZ+R is determined;

(d) a vector v_m is formed as the dot product of the measured spectrum of the sample under consideration, x_u, with each of the columns of the correction matrix U_m, v_m =x_u^t U_m ;

(e) a corrected spectrum x_c is formed as a result of orthogonalising x_u with respect to U_m, x_c =x_u -U_m v_m^t ;

(f) the scores for the corrected spectrum are calculated as v=x_c^t UΣ

^-1 where x_c^t is the transpose of x_c ;

(g) the measurement process signals are calculated as r=v_m -vZ; and

(h) the magnitude of the elements of r are compared with the range of values of R, whereby a significant difference indicates a significant change in the measurement process data.

12. A method as claimed in claim 1 for which matrix X is to be corrected only for the effect of baseline variations, wherein said baseline variations are modelled by a set of orthogonal frequency (or wavelength) dependent polynomials which form said matrix U_m of dimension f by m where m is the order of the polynomials and the columns of U_m are orthogonal polynomials.

13. A method as claimed in claim 12, wherein said orthogonal polynomials are Legendre polynomials.

14. A method as claimed in claim 1 for which matrix X is to be corrected only for the effect of chemical compounds which do not form part of the sample whose unknown property data, composition data, or property and composition data is to be determined (hereafter "ex-sample chemical compounds"), wherein the spectra that form the columns of U_m are orthogonal vectors that are representatives of those ex-sample chemical compounds.

15. A method as claimed in claim 10, wherein step (ic) is performed by a Gram-Schmidt orthogonalization procedure.

16. A method as claimed in claim 10, wherein the ex-sample spectrum is that of water vapor.

17. A method as claimed in claim 10, wherein the ex-sample spectrum is that of carbon dioxide vapor.

18. A method as claimed in claim 10, where the ex-sample spectra are those of water vapor and carbon dioxide vapor.

19. A method as claimed in claim 10, wherein in step (id), a singular value decomposition of matrix X_s '"'"' is used to generate the orthogonal correction spectra Us, the first s'"'"' terms of which correspond to the number of different types of ex-sample spectral interferences being modelled are retained while any remaining terms are eliminated, and the resulting matrix U_s is combined with matrix U_p in step (ie) to form matrix U_m.

20. Apparatus for estimating unknown property and/or compositional data of a sample, comprising:

a spectrometer for generating the spectrum of a plurality of calibration samples n having known properties composition, or properties and composition c, and the spectrum of a sample under consideration having unknown property and/or composition data which is to be estimated; and

computer means arranged to receive the measured spectrum data from the spectrometer and operable

(i) in a data correction mode to perform, in conjunction with an operator, steps (i) to (iii) of claim 1;

(ii) in a storing mode to store the c property data, composition data or property and composition data for each of the n calibration samples to form the matrix Y of dimension n by c (c≧

1);

(iii) in a model building mode to determine, in conjunction with the operator, a predictive model according to step (v) of claim 1;

(iv) in a measurement mode to perform step (vi) of claim 12; and

(v) in a prediction mode to perform step (vii) of claim 12, in order to estimate the unknown property data, composition data, or property and composition data of the sample under consideration according to the determined predictive model correlating the elements of matrix Y to the corrected spectra matrix X_c resulting from the orthogonalization of matrix X with respect to the correction matrix U_m.