Method for Development of Independent Multivariate Calibration Models

• US 20080034025A1
• Filed: 07/04/2005
• Published: 02/07/2008
• Est. Priority Date: 07/27/2004
• Status: Abandoned Application

• Current Assignee: Aleksandr Shamrai, Konstantin Zharinov, Vladimir Zubkov
• Original Assignee: Aleksandr Shamrai, Konstantin Zharinov, Vladimir Zubkov
• Inventors: Zubkov, Vladimir, Zharinov, Konstantin, Shamrai, Aleksandr
• Application Number:
  US11/572,626
• Publication Number:
  US 20080034025A1
• Time in Patent Office: Days
• US Class Current: 708/403
• CPC Class Codes: G01D 18/00 Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00

Abstract

A method is provided for development of independent multivariate calibration models used for determination of secondary properties of a sample based on results of measurements of plurality of primary properties of the sample on target instrument. The method allows development of an independent calibration model for the target instrument regarding results of measurements of plurality of primary properties for sufficient amount of calibration samples with known secondary properties at the reference instrument. In this method, a set of transfer samples are selected, and a plurality of measurements of primary properties are made for each transfer sample using both reference and target instruments. Then relationships of transformation are generated from comparison of the target and reference instruments responses to the transfer samples. These relationships capable to perform a multivariate estimations of the target instrument response for each sample from the calibration set based on the reference instrument responses for those samples. The estimated responses of the target instrument for calibration samples are then used to calculate calibration constants in relationships relating estimated primary properties of each calibration sample to known corresponding secondary properties. Thus, an independent calibration model is developed at the target instrument, which allows extending of the calibration sample set and using outlier statistics for detection of outlier samples in the calibration set and for identification of unknown sample that can be analyzed.

Claims

1. A method for the development of independent multivariate calibration models comprising the following steps:

selecting a set of calibration samples with known secondary properties;

measuring the primary properties for each calibration sample on a reference instrument;

converting the measurement results of the primary properties for each calibration sample obtained on the reference instrument to a structure which is equivalent to measurement results for the calibration samples obtained on a calibratable instrument, wherein the conversion is made using correlations of calibration transfer;

comparing the measurement results of the primary properties for each calibration sample measured on the reference instrument which were converted to a structure being equivalent to measurement results obtained on a calibrated instrument with the known secondary properties of the calibration samples;

generating an independent multivariate calibration model by methods of multivariate regression analysis, wherein the independent multivariate calibration model expresses secondary properties of the samples by means of the primary properties measured on the calibrated instrument,

characterized in that

the mathematical correlations of calibration transfer are determined by;

selecting the set of samples for calibration transfer;

measuring the primary properties of each sample for calibration transfer on both the reference instrument and the calibrated instrument;

comparing the results of measurements of the primary properties of the samples for calibration transfer obtained on the reference instrument with the results of measurements of the primary properties of the same samples obtained on the calibrated instrument, using thereby the methods of multivariate regression analysis;

wherein the independent multivariate calibration model is selected using quantitative parameters of calibration verification.

2. The method of claim 1, wherein before correlating the results of measurements of the primary properties of the transfer samples on the both instruments, these results are subjected to normalization procedure to reveal a difference in the calibrated and reference instrument responses, and the results of measurements of the primary properties of the calibration samples on the reference instrument are subjected to the same normalization procedure before their transformation to a form equivalent to the calibrated instrument responses for these samples.

3. The method of claim 1, wherein the samples which define the range of possible differences in results of measurements of the primary properties on both the calibrated and reference instruments are selected in the set of transfer samples.

4. The method of claim 1, wherein the samples with known secondary properties are selected into the set of transfer samples and changes in the secondary properties of the selected samples define the range of possible differences in analyzed secondary properties.

5. The method of claim 1, wherein if the validation procedure shows inconsistent model accuracy, the estimated calibrated instrument responses for each calibration sample are analyzed by outlier statistics for the detection of outlier samples, and the detected outlier samples are discarded from the set of calibration samples before the development of an updated independent calibration model.

6. The method of claim 1, wherein if the validation procedure shows inconsistent model accuracy, the estimated calibrated instrument responses for each calibration sample are supplemented by results of measurements of the primary properties of additional calibration samples with known secondary properties on the target instrument whereby the set of calibration samples is expanded, and an updated independent calibration model is developed by processing the target instrument responses for the expanded set of calibration samples.

7. The method of claim 6, wherein the calibrated instrument responses for an expanded set of calibration samples are analyzed by outlier statistics for the detection of outlier samples, and the detected outlier samples are discarded from the expanded set of calibration samples before the development of updated independent calibration model.

8. The method of claim 6, wherein the responses of any instrument (including a reference instrument) for an expanded set of calibration samples are estimated using transformation relationships obtained after measurements of the transfer samples on this instrument and correlating the results of measurements of the primary properties of transfer samples on this instrument with results of measurements of the same samples on the target instrument, and then a new independent calibration model is developed at this instrument by processing the estimated responses of this instrument for the expanded calibration set by methods of regression analysis so that a number of calibration constants in the relationships between the primary properties and the secondary properties are generated;

in this case, the target instrument stands for a new reference instrument and the other instrument stands for a new target instrument.

9. The method of claim 1, wherein if the validation procedure shows inconsistent model accuracy, the estimated calibrated instrument responses for each calibration sample and (or) the corresponding known secondary properties of the calibration samples are subjected to normalization procedure before the development of an updated independent calibration model.

10. The method of claim 6, wherein if the validation procedure shows inconsistent model accuracy, the calibrated instrument responses for an expanded set of calibration samples and (or) the corresponding known secondary properties of the calibration samples are subjected to normalization procedure before the development of an updated independent calibration model.

11. The method of claim 1, wherein the reference and calibrated instruments are spectrometers and the measured primary properties are the light absorption, reflection or scattering at different wavelengths.

12. The method of claim 11, wherein the reference and target spectrometers are based on the principles of Fourier Transform spectroscopy and a wavelength shift is neglected in the transformation relationships for estimating the target instrument response for each calibration sample from of the reference instrument response for this sample.