INSTITUTE OF SPECTROSCOPY RUSSIAN ACADEMY OF SCIENCES 142090, ISAN, Troitsk city, Moscow region, Russia, Tel. 7(495)334-0235, fax 7(495)334-0886 [email protected].

INSTITUTE OF SPECTROSCOPY

RUSSIAN ACADEMY OF SCIENCES

142090, ISAN, Troitsk city, Moscow region, Russia,Tel. 7(495)334-0235, fax 7(495)334-0886

[email protected]

VI Winter Symposium on Chemometrics February 18—22 , 2008, Kazan, Russia

The study of dispersed systems using NIR spectroscopy and PLS regression technique

A. Kalinin, А. Potapov, S. Sadovsky

Multi-component calibration (construction of calibration models) a spectrometer for the quantitative analysis of multiphase materials includes:

1. -choice of a spectral method and spectral “hardware”- spectrometers. This defined by the physical and chemical properties of object and by a specific task of definition;

2. -creation or accumulation of the reference sample sets that meets well known requirements (on representativeness, density of filling of an analyzed range, on absence of significant correlation between components of a mix, etc.);

3. -construction and testing of calibration models, its optimization. A choice and adaptation of tools for preprocessing of spectra and of the regression analysis.

For the last step we have started up using the PLS program Quant 2 OPUS,

manufactured by BRUKER OPTIC Ltd. But later it was unsatisfactory because of:

1) labour inputs for loading of greater sets of spectra and reference data,2) has no some important intermediate interfaces,3) it is inaccessible to modernization for current problems,4) too expensive for the use in structure of our devices.

PLS program ISCAP had been developed, tested by comparing with OPUS results, and certified at Federal Inspection on Industry Property, Russian Patent Committee.

Now we use it for PLS calibration and prediction execution with our spectrometers and modify it for current problems.

WHAT’S TO DO:The leaders of Milk Institute of RAAS formulate for us the next tasks of dairy

quality control :

1. Fat, protein and lactose content prediction for raw milk. It’s obviously the most complex problem concerning of variability of physical and chemical structure of named object,

2. Definition of the nutritional value (fat, protein, lactose, saccharose, a dairy acid, calcium, cholesterol, etc.), in dairy produce and intermediate fractions (sour cream, concentrated milk, powder).

3. Recognition of fakes of dairy fats and proteins in dairy produce or in the intermediate fractions.

The special tasks are:

- to adapt the portable IR spectrometers for a prediction of quality

parameters of Russian dairy;

- to reduce cost of operational expenses.

These problems were solved in several countries by well known producers, but (as can be seen at every year Milk Forums at Moscow) the applications of their instruments for

Russian milk prediction isn’t successful till now.

Methods:NIR spectroscopy at transflectance and diffuse reflection modes,

PLS regression,

Laser correlation spectroscopy;

Dispersions as objects for constituent and parameter prediction:

-liquid and powdered milk,

-hydrated reversed micelles (water-aerosol OT-octan)

-water-ethanol-gasoline mixtures.

Now we will show the selected results dealing with:

Calibration model with absorption spectra of 34 milk mixtures collected with spectrophotometer MC-75, described in:

A.V.Kalinin, V.N.Krasheninnikov, V.M.Krivtsun, S.V.Sadovsky, in: Book of abstracts, International Congress on Analytical Sciences, 25-30 June, M., 1, 157, (2006).

Pre-production spectrometer BIKAN-K operating in transflectance mode for fat, protein, etc. prediction in liquid dairy

Characteristics: • - Spectral range 800-

1070 nm;• - The sizes

31X22X12 сm;• - Weight 4,6 kg;• - Time of definition

of parameters in 10 samples - 6 minutes;

• - A block design, simplicity of service;

• - An opportunity of expansion of the basic models (for various a component, ranges of definition, products)

A.Kalinin, V.Krasheninnikov, V.Krivtsun, S.Sadovsky, H.Denisovich, H.Yurova, JNIRS, Proceedings of NIR2007, accepted

65

1-st der.

cm-1

cm-1

Diffuse reflection spectrometer BIKAN-CP for fat, protein, etc. prediction in

powder on the basis of polychromator with a liner InGaAs-detector;

Key characteristics: • - spectral range 1050-

1650 nm,• - thermoresisted

(<400Co) opto-fiber probe,

• dimensions 31X22X12 sm,

• weight of 4,5 kg;• - time of measurement

is 10 sec a sample; • - modular design,

simplicity of service; • - an opportunity of

expansion of bank calibration models (for defined components, their ranges, products, modes of measurement)

A.Kalinin, V.Krasheninnikov, V.Krivtsun, S.Sadovsky, H.Denisovich, H.Yurova, JNIRS, Proceedings of NIR2007, accepted

Application of a portable near infra-red spectrometer of diffuse reflection БИКАН-СР with a high-temperature probe for the multicomponent analysis of damp

powders and dispersed systems with uncontrollable temperature

How to collect the diffuse reflection spectra of milk powder located in thermogravimetric moisture analyzer (TGA) with uncorrelated moisture and

temperature value

The set-up for spectra and moisture and temperature data acquisition

A.Kalinin, V.Krasheninnikov, V.Krivtsun, S.Sadovsky, H.Denisovich, H.Yurova, JNIRS, Proceedings of NIR2007, accepted

nm

laser correlation spectroscopy of scattered light

H e-N e л азер

Ф Э У

Ф Э УК оррел ятор

П оворотное устройство

К ю ветаТерм остат

Лебедев А.Д. и др. Лазерная корреляционная спектроскопия в биологии. Киев, 1987.

A compact photon-correlation spectrometer enables one to make absolute measurements of the sizes and evaluate the shape of particles suspended in liquids in the range from 0.001 to 10 μm. The measurements are fast, lasting usually from seconds to several minutes.

rotator

laser correlation spectroscopy of scattered light

Curve 1 – theoretical value for micelle radius r, Curve 2 – the difference ΔR=Rh-r of experimental and theoretical value for micelle radius,Curve 3 – parameter P defining the nonsphericity of micelle shape (Р=0 –for sphere)

Potapov A.V., Saletsky A.M. Laser Physics Letters, 2005, v.2, №10, р.476-480

The evaluation of nonsphericity

R=.15w+.4

water-AOT-octane dispersion – the way for higher the sensitivity of NIR spectroscopy on water content

prediction

1.2nm

There was prepared 10 samples with uncorrelated water and AOT mass fractions and was measured the size distributions with the use of LCS. The result were:- The scattered light level was sufficient to measure size distribution for 6 samples with w>20,- All distributions contained one single peak, and the value of average radius Rav shifted with w as R=(1.5w+.4+1.2) nm with the error not more than 10%.

Potapov A.V., Saletsky A.M. Laser Physics Letters, 2005, v.2, №10, р.476-480

AOT molecule

cavity filled by water

This result gave us the possibility to calculate the concentration of micelles for each dispersion and to build calibration on it and on diffuse reflection spectra collected with BIKAN-CP spectrometer.

AOT- aerosol OT – dioctylsulfocuccein sodium – surface-active substance.It forms the micelles with water encapsulated in it of shown below structure if mass fractions of water and AOT content are related with “ratio of aquation”,w in the well known range: 20<w<70

Calibrasion model c_p_Raw10 for water prediction with the use of diffuse reflection NIR spectra

constituent N of factors R RMSECV Range of var. Part_conc 1 99.80 0.19 0.58 - 12.64, r.u.

Water mass. fr. 4 77.59 0.25 0.02 - 1.26 w%

Number of points in spectra - 253 Spectral range: 1065 - 1650 nm Number of samples for calibration 10 Pretreatments - raw

Cross validationConc. of part.

N of factors -1

Cross validationWater mass. fr.N of factors-4

Calibration model built on diffuse reflection spectra collected with BIKAN-CP spectrometer of reversed micells in water-AOT-octane dispersions

This is the way for higher the sensitivity of NIR spectroscopy on water content prediction

The study of water-ethanol-gasoline dispersion was based on the model of water-ethanol scatters accordingly to A.M.Saletskii and others, Phys. Rev. Letters,1999, v.12, №1, р.124-

127

-The LCS size distribution typically had two, three and more peaks,- There is no any correlation between Rav and mass fractions of water and ethanol, So, it was not possible to define the value of scatter concentration and to look for correlation of spectra and scatter concentration variation

We prepared 30 mixes, measured size distributions with the LCS and NIR spectra with the use of BIKAN-CP

Calibration model is trivial:Ethanol content may be predicted,but not the water one.

rav = 208 nm

rav = 8 nm

Distribution on the sizes of associations of casein - a) and globulin–b) molecules in milk

Calibration model with unsatisfactory correlation of transflectance spectraand protein (globulin) content variations caused possibly by twenty fold less sizes of

globulin molecular associations comparing with the casein ones

a

b

The useful result is: the spectrometer BIKAN-K isn’t sensible to fakes with serum protein concentrate

Conclusions:

Automation of calibration model execution allows to consider at once a several models for each set of spectra and sample data, to compare their efficiency (reliability, complexity, an error, etc. qualities) and to estimate correctness of decisions on spectral technique and on calibration set design,

Laser correlation spectrometry of nano-dispersive phase is useful tool of radiation-matter interaction research.

Thank you for attention!

Aknowledgements:

- to ms Andrey Tretiyakov for preparation of dispersions,- to ms Alex Novosielov for operating with LSC,- to doctor Victor Krasheninnikov for operating with spectrometers and useful discussions