Thesis (MScAgric)--Stellenbosch University, 2013. / ENGLISH ABSTRACT: The WineScan FT120 is widely used in wine laboratories across South Africa. The WineScan
FT120 uses Fourier transform infrared (FT-IR) spectroscopy with multivariate data analysis to
correlate spectra with chemical compositional data. Ready-to-use, commercially available
calibration models for a FT-IR spectroscopy instrument are an advantage for unskilled users
and routine analysis. Introducing spirit products to this technology introduced new interferences,
which necessitated vastly different calibrations models to compensate for the changes.
Accuracy, precision and ruggedness of the reference methods validated during method
validation, verified the suitability of the reference methods used to quantify the parameters in
question before calibration model building was attempted.
Various principal component analysis (PCA) were performed prior to the calibration step
with the aim to identify outliers and inspect groupings. PCA models could identify samples with
atypical spectra and differentiate between product types.
Two tactics regarding data sets for calibration set-up was experimented with, all the
products together and calibration models per product. Partial least squares (PLS) regression
was used to establish the calibration models for ethanol, density, obscuration and colour. With
all the calibration models, the calibration models based on the product specific data sets,
achieved better predicting statistics. The best performing ethanol calibration models achieved
Residual mean square error of prediction (RMSEP) = 0.038 to 0.106 %v/v and showed
significant improvement on previously reported prediction errors by Lachenmeier (2007). The
results for the density calibration showed a similar trend, with the product specific calibration
models outperforming the calibration model when all samples were included into one calibration
model. This study produced novel results for quantification of obscuration (RMSEP = 0.10 and
0.09 in blended brandies and potstill brandies, respectively) and colour (RMSEP < 2.286 gold
units) of brandies and whiskies. The correlation coefficients (R²) between true and predicted
values, for the four parameters tested, indicated good to excellent precision (0.8 < R² < 1.0).
Minimising the variation between the samples of the data set, gave more accurate regression
statistics, but this resulted in a lower residual predictive deviation (RPD) value (< 5) that
indicated models were not suitable for quantification. Adding more samples per product will add
more variability into a data set per product, increase the SD and result in an increase in the
RPD. The results pave the way for the development of calibration models for the quantification
of other parameters for specific products.
Following the groupings of product types, further classifications of brandy brands were
investigated. PCA plots showed clear separation between potstill brandies and blended
brandies and some degree of clustering between some of the blended brands was observed.
Classification of brandies were investigated using the Soft Independent Modeling of Class
Analogy (SIMCA) approach resulting in a total correct classification rates between 81.25% and
100% for the various brandy brands. These preliminary results were very promising and
highlight the potential of using FT-IR spectroscopy and multivariate classification techniques as
a tool for rapid quality control and authentication of brandy brands.
Using this work as base for further classification projects, this could be of great benefit to
the alcoholic beverage industry of South Africa. Future work will involve the development of a
database comprised of more products guaranteed authentic to expand the discriminating
options. The results suggest FT-IR spectroscopy could be useful in authentication studies. / AFRIKAANSE OPSOMMING: Die WineScan FT120 is ‘n algemeen gebruikte instrument regoor Suid-Afrika. Die WineScan
FT120 gebruik Fourier-transformasie-infrarooi (FT-IR) spektroskopie tesame met
multiveranderlike statistiese metodes om spektra te korreleer met chemiese samestellingsdata.
Die kommersieël beskikbare kalibrasiemodelle vir die FT-IR spektroskopie-instrument is ‘n
voordeel vir onbedrewe gebruikers en roetine ontleding. Blootstelling van spiritusprodukte aan
die tegnologie, het nuwe hindernisse bekend gestel en dus is verskillende kalibrasiemodelle
genoodsaak om hiervoor te kompenseer.
Akkuraatheid, presiesheid en ruheid van die verwysingsmetodes is geëvalueer tydens
metodevalidasie. Die verwysingsmetodes is geskik verklaar vir die konstruksie van die
kalibrasiemodel met geverifieërde akkurate verwysingsresultate.
Verskeie multiveranderlike hoofkomponentanalise (MVK) was uitgevoer voor die kalibrasiestap
met die doel om uitskieters te identifiseer en groeperings te inspekteer. MVK modelle kon
monsters met atipiese spektra identifiseer en onderskei tussen verskillende produk tipes.
Twee taktieke aangaande datastelsamestelling is getoets tydens kalibrasiemodel-opstelling,
al die produkte saam en kalibrasiemodelle per produk soos met die MVK aangedui. Parsiële
kleinste kwadraat (PKK)- regressie is gebruik vir die opstel van die kalibrasiemodelle vir etanol,
digtheid, obskurasie en kleur. Met al die kalibrasiemodelle het die produk spesifieke
kalibrasiemodelle beter regressiestatistiek gelewer. Die beste presterende etanol
kalibrasiemodelle het ‘n standaardvoorspellingsfout (SVF) = 0.038 tot 0.106 %v/v bereik en het
‘n beduidende verbetering getoon op vorige gerapporteerde studies op spiritusprodukte
(Lachenmeier, 2007). Die resultate vir die digtheidskalibrasiemodelle het ‘n eenderse tendens
getoon soos die etanol, met die produk spesifieke kalibrasiemodelle wat beter presteer het.
Hierdie studie was eerste in sy soort met die kalibrasiemodel vir obskurasie (SVF = 0.10 en 0.09
in gemengde brandewyne en potketel brandewyne, onderskeidelik) en kleur (SVF < 2.286 goud
eenhede) van brandewyne en whiskies. Die bepalingskoëffisiënt (R²) vir die vier parameters, dui
op goeie tot uitstekende presiesheid (0.8 < R² < 1.0). Vermindering van die variasie tussen die
monsters in die datastel, het meer akkurate regressiestatistiek teweeg gebring, maar ‘n laer
relatiewe voorspellingsafwyking (RVA) waarde (<5) tot gevolg gehad wat aan dui dat hierdie
modelle nie geskik is vir sifting of kwantifisering nie. Die byvoeging van meer monsters per
produk sal meer verskeidenheid in die datastel per produk bring, wat dan die standaardafwyking
sal laat toeneem en uiteindelik die RVA laat toeneem. Die resultate het die fondasie gelê vir die
ontwikkeling van kalibrasiemodelle vir die kwantifisering van ander parameters vir spesifieke
produkte.
As opvolg tot die groeperings van die produk tipe, waargeneem in die MVK modelle, was
klassifikasie van brandewyn handelsmerke ondersoek. MVK modelle het duidelike skeiding
gewys tussen potketel en gemengde brandewyne en tot ‘n sekere mate groepering tussen
handelsmerke. Klassifikasie van brandewyne was ondersoek met behulp van the Soft
Independent Modeling of Class Analogy (SIMCA) met die resultaat van ‘n totale korrekte
klassifikasiekoers van tussen 81.25% en 100% vir die verskeie brandewyn handelsmerke.
Hierdie voorlopige resultate toon belowend en beklemtoon die potensiaal van FT-IR
spektroskopie en chemometrics tegnieke as toerusting vir die vinnige kwaliteitskontrole en
egtheid van brandewyn handelsmerke studies.
Met hierdie werk as basis vir verdere klassifikasie projekte, kan dit ‘n groot aanwins wees
tot die alkoholiese drank industrie van Suid-Afrika. Toekomstige werk sal insluit die ontwikkeling
van ‘n databasis saamgestel met meer gewaarborgde egte produkte om die klassifikasie uit te
brei.
| Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:sun/oai:scholar.sun.ac.za:10019.1/85734 |
| Date | 12 1900 |
| Creators | Kleintjes, Tania Victoria |
| Contributors | Lambrechts, Marius, Nieuwoudt, Helene, Stellenbosch University. Faculty of AgriSciences. Dept. of Viticulture and Oenology. |
| Publisher | Stellenbosch : Stellenbosch University |
| Source Sets | South African National ETD Portal |
| Language | en_ZA |
| Detected Language | English |
| Format | 99 p. : ill. |
| Rights | Stellenbosch University |
Page generated in 0.0076 seconds