Yes / Polymer nanocomposites are usually characterized using various methods, such as small angle X-ray diffraction (XRD) or transmission electron microscopy, to gain insights into the morphology of the material. The disadvantages of these common characterization methods are that they are expensive and time consuming in terms of sample preparation and testing. In this work, near infrared spectroscopy (NIR) spectroscopy is used to characterize nanocomposites produced using a unique twin-screw mini-mixer, which is able to replicate, at ~25 g scale, the same mixing quality as in larger scale twin screw extruders. We correlated the results of X-ray diffraction, transmission electron microscopy, G′ and G″ from rotational rheology, Young’s modulus, and tensile strength with those of NIR spectroscopy. Our work has demonstrated that NIR-technology is suitable for quantitative characterization of such properties. Furthermore, the results are very promising regarding the fact that the NIR probe can be installed in a nanocomposite-processing twin screw extruder to measure inline and in real time, and could be used to help optimize the compounding process for increased quality, consistency, and enhanced product properties

This diploma thesis provides summary information about using of near infrared spectroscopy for monitoring of high oleic acid content in sunflower seed. The calibration for high oleic sunflower seed was performed by using calibration model PLS - „Partial least squares. Gas chromatography was used as a reference method. Based on this knowledge thirty samples of high oleic sunflower seeds were measured and evaluated on this calibration. The last part of diploma thesis consist of comparison of oleic acid results measured by near infrared spectroscopy and gas chromatography used as reference method, and discussion about the measurement accuracy and evaluation of suitability of using these methods.

Thesis (PhD)--Stellenbosch University, 2013. / ENGLISH ABSTRACT: Concrete evidence of dryland salinity was observed in the Berg River catchment in the Western Cape Province of South Africa. Soil salinization is a global land degradation hazard that negatively affects the productivity of soils. Timely and accurate detection of soil salinity is crucial for soil salinity monitoring and mitigation. It would be restrictive in terms of costs to use traditional wet chemistry methods to detect and monitor soil salinity in the entire Berg River catchment. The goal of this study was to investigate less tedious, accurate and cost effective techniques for better monitoring. Firstly, hyperspectral remote sensing (HRS) techniques that can best predict electrical conductivity (EC) in the soil using individual bands, a unique normalized difference soil salinity index (NDSI), partial least squares regression (PLSR) and bagging PLSR were investigated. Spectral reflectance of dry soil samples was measured using an analytical spectral device FieldSpec spectrometer in a darkroom. Soil salinity predictive models were computed using a training dataset (n = 63). An independent validation dataset (n = 32) was used to validate the models. Also, field-based regression predictive models for EC, pH, soluble Ca, Mg, Na, Cl and SO4 were developed using soil samples (n = 23) collected in the Sandspruit catchment. These soil samples were not ground or sieved and the spectra were measured using the sun as a source of energy to emulate field conditions. Secondly, the value of NIR spectroscopy for the prediction of EC, pH, soluble Ca, Mg, Na, Cl, and SO4 was evaluated using 49 soil samples. Spectral reflectance of dry soil samples was measured using the Bruker multipurpose analyser spectrometer. “Leave one out” cross validation (LOOCV) was used to calibrate PLSR predictive models for EC, pH, soluble Ca, Mg, Na, Cl, and SO4. The models were validated using R2, root mean square error of cross validation (RMSECV), ratio of prediction to deviation (RPD) and the ratio of prediction to interquartile distance (RPIQ). Thirdly, owing to the suitability of land components to map soil properties, the value of digital elevation models (DEMs) to delineate accurate land components was investigated. Land components extracted from the second version of the 30-m advanced spaceborne thermal emission and reflection radiometer global DEM (ASTER GDEM2), the 90-m shuttle radar topography mission DEM (SRTM DEM), two versions of the 5-m Stellenbosch University DEMs (SUDEM L1 and L2) and a 5-m DEM (GEOEYE DEM) derived from GeoEye stereo-images were compared. Land components were delineated using the slope gradient and aspect derivatives of each DEM. The land components were visually inspected and quantitatively analysed using the slope gradient standard deviation measure and the mean slope gradient local variance ratio for accuracy. Fourthly, the spatial accuracy of hydrological parameters (streamlines and catchment boundaries) delineated from the 5-m resolution SUDEM (L1 and L2), the 30-m ASTER GDEM2 and the 90-m SRTM was evaluated. Reference catchment boundary and streamlines were generated from the 1.5-m GEOEYE DEM. Catchment boundaries and streamlines were extracted from the DEMs using the Arc Hydro module for ArcGIS. Visual inspection, correctness index, a new Euclidean distance index and figure of merit index were used to validate the results. Finally, the value of terrain attributes to model soil salinity based on the EC of the soil and groundwater was investigated. Soil salinity regression predictive models were developed using CurveExpert software. In addition, stepwise multiple linear regression soil salinity predictive models based on annual evapotranspiration, the aridity index and terrain attributes were developed using Statgraphics software. The models were validated using R2, standard error and correlation coefficients. The models were also independently validated using groundwater hydro-census data covering the Sandspruit catchment. This study found that good predictions of soil salinity based on bagging PLSR using first derivative reflectance (R2 = 0.85), PLSR using untransformed reflectance (R2 = 0.70), a unique NDSI (R2 = 0.65) and the untransformed individual band at 2257 nm (R2 = 0.60) predictive models were achieved. Furthermore, it was established that reliable predictions of EC, pH, soluble Ca, Mg, Na, Cl and SO4 in the field are possible using first derivative reflectance. The R2 for EC, pH, soluble Ca, Mg, Na, Cl and SO4 predictive models are 0.85, 0.50, 0.65, 0.84, 0.79, 0.81 and 0.58 respectively. Regarding NIR spectroscopy, validation R2 for all the PLSR predictive models ranged from 0.62 to 0.87. RPD values were greater than 1.5 for all the models and RMSECV ranged from 0.22 to 0.51. This study affirmed that NIR spectroscopy has the potential to be used as a quick, reliable and less expensive method for evaluating salt-affected soils. As regards hydrological parameters, the study concluded that valuable hydrological parameters can be derived from DEMs. A new Euclidean distance ratio was proved to be a reliable tool to compare raster data sets. Regarding land components, it was concluded that higher resolution DEMs are required for delineating meaningful land components. It seems probable that land components may improve salinity modelling using hydrological modelling and that they can be integrated with other data sets to map soil salinity more accurately at catchment level. In the case of terrain attributes, the study established that promising soil salinity predictions could be made based on slope, elevation, evapotranspiration and terrain wetness index (TWI). Stepwise multiple linear regressions soil salinity predictive model based on elevation, evapotranspiration and TWI yielded slightly more accurate prediction of soil salinity. Overall, the study showed that it is possible to enhance soil salinity monitoring using HRS, NIR spectroscopy, land components, hydrological parameters and terrain attributes. / AFRIKAANSE OPSOMMING: Konkrete bewyse van droëland sout is waargeneem in die Bergrivier opvanggebied in die Wes- Kaap van Suid-Afrika. Verbrakking van grond is 'n wêreldwye probleem wat ‘n negatiewe invloed op die produktiwiteit van grond kan hê. Tydige en akkurate herkenning van verandering in grond soutgehalte is ‘n noodsaaklike aksie vir voorkoming. Dit sou beperkend wees in terme van koste om konvensionele nat chemiese metodes te gebruik vir die opsporing en monitering daarvan in die hele Bergrivier opvanggebied. Die doel van hierdie studie was om ondersoek in te stel na minder tydsame, akkurate en koste-effektiewe tegnieke vir beter monitering. Eerstens, is hiperspektrale afstandswaarnemings (HRS) tegnieke wat die beste in staat is elektriese geleidingsvermoë (EG) in die grond te kan voorspel deur gebruik te maak van individuele bande, 'n unieke genormaliseerde grond soutindeks verskil (NDSI), parsiële kleinste kwadratiese regressie (PLSR) en afwyking in PLSR, is ondersoek. Spektrale reflektansie van droë grondmonsters is gemeet deur gebruik te maak van 'n spektrale analitiese toestel: FieldSpec spektrometer in 'n donkerkamer. Voorspellings modelle vir grond soutgehalte is bereken met behulp van 'n toets datastel (n = 63). 'n onafhanklike validasie datastel (n = 32) is gebruik om die modelle te evalueer. Daarbenewens is veld-gebaseerde regressie voorspellings modelle vir EG, pH oplosbare Ca, Mg, Na, Cl and SO4 ontwikkel deur gebruik te maak van grondmonsters (n = 23) versamel in the Sandpruit opvangsgebied. Hierdie grondmonsters is nie gemaal of gesif nie en die spectra is gemeet deur gebruik te maak van die son as ‘n bron van energie om veld toestande na te boots. Tweedens, is die waarde van NIR spektroskopie vir die voorspelling van die EG, pH, oplosbare Ca, Mg, Na, Cl, en SO4 met behulp van 49 grondmonsters geëvalueer. Spektrale reflektansie van droë grondmonsters is gemeet deur gebruik te maak van die Bruker NIR veeldoelige analiseerder . Kruisvalidering (LOOCV) is gebruik om PLSR voorspellings modelle vir EG, pH, oplosbare Ca, Mg, Na, Cl, en SO4 te kalibreer. Hierdie modelle is gevalideer: R2, wortel-gemiddelde-kwadraat fout kruisvalidering (RMSECV), verhouding van voorspellings afwyking (RPD) en die verhouding van die voorspelling se inter-kwartiel afstand (RPIQ). Derdens is land komponente gekarteer vanweë die nut daat van tov grondeienskappe, en die waarde van DEMs is ondersoek om akkurate land komponente af te baken. Land komponente uit die tweede weergawe van die 30 m gevorderde ruimte termiese emissie en refleksie radio globale DEM (ASTER GDEM2), die 90-m ruimtetuig radar topografie sending DEM (SRTM DEM), twee weergawes van die 5 m Universiteit van Stellenbosch DEMs (SUDEM L1 en L2) en 'n 5 m DEM (GEOEYE DEM) afgelei van GeoEye stereo-beelde, is vergelyk. Land komponente is afgebaken met behulp van helling, gradiënt en aspek afgeleides van elke DEM. Die land komponente is visueel geïnspekteer en kwantitatief ontleed met behulp van die helling gradiënt standaardafwyking te meet en die gemiddelde helling-gradiënt-plaaslike variansie verhouding vir akkuraatheid. Vierdens, is die ruimtelike akkuraatheid van hidrologiese parameters (stroomlyn en opvanggebied grense) geëvalueer soos afgelei vanaf die 5 m resolusie SUDEM (L1 en L2), die 30 m ASTER GDEM2 en die 90 m SRTM . Die verwysings opvanggebied grens en stroomlyn is gegenereer vanaf die 1,5-m GEOEYE DEM. Opvanggebied grense en stroomlyn uit die DEMs is bepaal deur gebruik te maak van die Arc Hydro module in ArcGIS. Visuele inspeksie, korrektheid indeks, 'n nuwe Euklidiese afstand indeks en die indikasie-van-meriete indeks is gebruik om die resultate te valideer. Laastens is die waarde van die terrein eienskappe om grond southalte te modeleer ondersoek, gebaseer op die EG van die grond en grondwater. Grond soutgehalte regressie voorspellings modelle is ontwikkel met behulp van CurveExpert sagteware. Verder, stapsgewyse meervoudige lineêre regressie grond soutgehalte voorspellings modelle gebaseer op jaarlikse evapotranspirasie, die dorheids indeks en terrein eienskappe is ontwikkel met behulp van Statgraphics sagteware. Die modelle is gevalideer deur gebruik te maak van R2, standaardfout en korrelasiekoëffisiënte. Die modelle is ook onafhanklik bekragtig deur die gebruik van grondwater hidro-sensus-data wat die Sandspruit opvanggebied insluit. Hierdie studie het bevind dat 'n goeie voorspelling van grond soutgehalte gebaseer op uitsak PLSR met behulp van eerste orde afgeleide reflektansie (R2 = 0,85), PLSR deur gebruik te maak van ongetransformeerde reflektansie (R2 = 0,70), 'n unieke NDSI (R2 = 0,65) en die ongetransformeerde individuele band op 2257 nm (R2 = 0,60) voorspellings modelle verkry is. Verder is vasgestel dat betroubare voorspellings van die EG, pH, oplosbare Ca, Mg, Na, Cl en SO4 in die veld moontlik is met behulp van eerste afgeleide reflektansie. Die R2 van EG, pH, oplosbare Ca, Mg, Na, Cl en SO4 is 0.85, 0.50, 0.65, 0.84, 0.79, 0.81 en 0.58 onderskeidelik. Ten opsigte van NIR spektroskopie het die validasie van R2 vir al die PLSR voorspellings modelle gewissel tussen 0,62-0,87. Die RPD waardes was groter as 1,5 vir al die modelle en RMSECV het gewissel tussen 0,22-0,51. Hierdie studie het bevestig dat NIR spektroskopie die potensiaal het om gebruik te word as 'n vinnige, betroubare en goedkoper metode vir die analise van soutgeaffekteerde gronde. T.o.v. hidrologiese parameters, het die studie tot die gevolgtrekking gekom dat waardevolle hidrologiese parameters afgelei kan word uit DEMs. 'n nuwe Euklidiese afstand verhouding is bevestig as 'n betroubare hulpmiddel om raster datastelle te vergelyk. Ten opsigte van grond komponente, is daar tot die gevolgtrekking gekom dat hoër resolusie DEMs nodig is vir die bepaling van sinvolle land komponente. Dit lyk waarskynlik dat die land komponent soutgehalte modellering hidrologiese modellering verbeter en dat hulle geïntegreer kan word met ander datastelle vir meer akkurate kaarte op opvangsgebied skaal. In die geval van die terrein eienskappe het, die studie vasgestel dat belowende grond soutgehalte voorspellings gemaak kan word gebaseer op helling, elevasie, evapotranspirasie en terrein natheid indeks (TWI). 'n stapsgewyse meervoudige lineêre regressie grond soutgehalte voorspellings model wat gebaseer is op elevasie, evapotranspirasie en TWI het effens meer akkurate voorspellings van die grond soutgehalte gelewer. In geheel gesien, het die studie getoon dat dit moontlik is om grond soutgehalte monitering te verbeter met behulp van HRS, NIR spektroskopie, land komponente, hidrologiese parameters en terrein eienskappe. / The Agricultural Research Council (ARC), Water Research Commission and the National Research Foundation for funding.

Computer technologies have revolutionised the processing of information and the search for knowledge. With the ever increasing computational power, it is becoming possible to tackle new data analysis applications as diverse as mining the Internet resources, analysing drugs effects on the organism or assisting wardens with autonomous video detection techniques. Fundamentally, the principle of any data analysis task is to fit a model which encodes well the dependencies (or patterns) present in the data. However, the difficulty is precisely to define such proper model when data are noisy, dependencies are highly stochastic and there is no simple physical rule to represent them. The aim of this work is to discuss the principles, the advantages and weaknesses of the probabilistic modelling framework for data analysis. The main idea of the framework is to model dispersion of data as well as uncertainty about the model itself by probability distributions. Three data analysis tasks are presented and for each of them the discussion is based on experimental results from real datasets. The first task considers the problem of linear subspaces identification. We show how one can replace a Gaussian noise model by a Student-t noise to make the identification more robust to atypical samples and still keep the learning procedure simple. The second task is about regression applied more specifically to near-infrared spectroscopy datasets. We show how spectra should be pre-processed before entering the regression model. We then analyse the validity of the Bayesian model selection principle for this application (and in particular within the Gaussian Process formulation) and compare this principle to the resampling selection scheme. The final task considered is Collaborative Filtering which is related to applications such as recommendation for e-commerce and text mining. This task is illustrative of the way how intuitive considerations can guide the design of the model and the choice of the probability distributions appearing in it. We compare the intuitive approach with a simpler matrix factorisation approach.

Soil compaction is a widely spread problem in agricultural soils that has negative agronomic and environmental impacts. The former may lead to poor crop growth and yield, whereas the latter may lead to poor hydraulic properties of soils, and high risk to flooding, soil erosion and degradation. Therefore, the elimination of soil compaction must be done on regular bases. One of the main parameters to quantify soil compaction is soil bulk density (BD). Mapping of within field variation in soil BD will be a main requirement for within field management of soil compaction. The aim of this research was to develop a new approach for the measurement of soil BD as an indicator of soil compaction. The research relies on the fusion of data from visible and near infrared spectroscopy (vis-NIRS), to measure soil gravimetric moisture content (ω), with frequency domain reflectometry (FDR) data to measure soil volumetric moisture content (θv). The values of the estimated ω and θv, for the same undisturbed soil samples were collected from selected locations, textures, soil moisture contents and land use systems to derive soil BD. A total of 1013 samples were collected from 32 sites in the England and Wales. Two calibration techniques for vis-NIRS were evaluated, namely, partial least squares regression (PLSR) and artificial neural networks (ANN). ThetaProbe calibration was performed using the general formula (GF), soil specific calibration (SSC), the output voltage (OV) and artificial neural networks (ANN). ANN analyses for both ω and θv properties were based either on a single input variable or multiple input variables (data fusion). Effects of texture, moisture content, and land use on the prediction accuracy on ω, θv and BD were evaluated to arrive at the best experimental conditions for the measurement of BD with the proposed new system. A prototype was developed and tested under laboratory conditions and implemented in-situ for mapping of ω, θv and BD. When using the entire dataset (general data set), results proved that high measurement accuracy can be obtained for ω and θv with PLSR and the best performing traditional calibration method of the ThetaProbe with R2 values of 0.91 and 0.97, and root mean square error of prediction (RMSEp) of 0.027 g g-1 and 0.019 cm3 cm-3, respectively. However, the ANN – data fusion method resulted in improved accuracy (R2 = 0.98 and RMSEp = 0.014 g g-1 and 0.015 cm3 cm-3, respectively). This data fusion approach gave the best accuracy for BD assessment when only vis-NIRS spectra and ThetaProbe V were used as an input data (R2 = 0.81 and RMSEp = 0.095 g cm-3). The moisture level (L) impact on BD prediction revealed that the accuracy improved with soil moisture increasing, with RMSEp values of 0.081, 0.068 and 0.061 g cm-3, for average ω of 0.11, 0.20 and 0.28 g g-1, respectively. The influence of soil texture was discussed in relation with the clay content in %. It was found that clay positively affected vis-NIRS accuracy for ω measurement and no obvious impact on the dielectric sensor readings was observed, hence, no clear influence of the soil textures on the accuracy of BD prediction. But, RMSEp values of BD assessment ranged from 0.046 to 0.115 g cm-3. The land use effect of BD prediction showed measurement of grassland soils are more accurate compared to arable land soils, with RMSEp values of 0.083 and 0.097 g cm-3, respectively. The prototype measuring system showed moderate accuracy during the laboratory test and encouraging precision of measuring soil BD in the field test, with RMSEp of 0.077 and 0.104 g cm-3 of measurement for arable land and grassland soils, respectively. Further development of the prototype measuring system expected to improve prediction accuracy of soil BD. It can be concluded that BD can be measured accurately by combining the vis-NIRS and FDR techniques based on an ANN-data fusion approach.

A metodologia da espectroscopia no infravermelho próximo (NIRS Near Infrared Spectroscopy), recentemente empregada em várias áreas da ciência do solo associada à quimiometria, está sendo utilizada na quantificação de atributos químicos e físicos do solo. Esta técnica rápida, não destrutiva, reprodutiva e de baixo custo fundamenta-se na medida da intensidade de absorção de radiação eletromagnética na região do infravermelho próximo. De um modo geral, a NIRS tem se mostrado uma ferramenta válida para quantificar o teor de carbono (C) em amostras de solo. A importância da determinação do teor de C no solo reside no fato de representar a fixação do CO2 atmosférico na forma de matéria orgânica do solo (MOS), um compartimento chave do ciclo global deste elemento. O manejo sustentável dos agrossistemas deve assegurar a manutenção dos teores de carbono através da restituição de matéria orgânica ao solo. No caso da cultura de cana de açúcar, a substituição do sistema de colheita de manual com prévia queima da palhada (cana queimada) para mecanizada (cana crua) em que a palhada permanece sobre o solo, favorece o acúmulo e a diferenciação do teor de C no solo. O presente trabalho teve como objetivo comparar a metodologia NIRS, associada à quimiometria, com o método de referência tradicional da combustão a seco na quantificação do teor de C em amostras de solo provenientes do agroossistema cana-de-açúcar. Foram analisadas amostras de solo provenientes de três situações: cana queimada, cana crua e uma área nativa, totalizando 450 amostras, apresentando respectivamente teores médios de C de 18,30, 22,36 e 24,72 g kg-1. Modelos de calibração foram ajustados utilizando PLS nos dados espectrais, submetidos a 1ª derivada suavizados com Savitz Golay alisados com janela 15. As amostras foram aleatorizadas e divididas em 3 partes iguais, sendo 2/3 das 450 amostras para a calibração e validação cruzada e 1/3 para validação externa. Os parâmetros quimiométricos para a avaliação da qualidade do modelo escolhido foram: RMSEC, RMSECV, RMSEP, R2, BIAS, CV% e RPD. Os resultados de RMSEC> 1,93, RMSEP > 1,90, satisfatórios R2 0,94. Na área nativa obteve-se RMSEP 1,53 e R2 0,95, e na área de cana queimada RMSEP 1,90 e R2 0,50. Ao comparar as técnicas, de referência e NIRS, demonstrou que a técnica de referência tem SD 0,12 e CV% 0,65 e a técnica NIRS SD 0,18 e CV% 0,18, provando a eficiência do emprego da NIRS para a determinação de C no solo / The methodology of near infrared spectroscopy, recently used in several areas of soil science associated with chemometrics, has been used in the quantification of chemical and physical soil attributes. This rapid, nondestructive, reproductive and low cost technique is based on the measurement of the intensity of the electromagnetic radiation absorption in the near infrared region. In general, the NIRS has been proved as an effective tool to quantify the carbon (C) content in soil samples. The importance this determination is that it represents the fixation of atmospheric CO2 as soil organic matter (SOM), an important compartment of the global C cycle. The sustainable management of agro-systems must assure the maintenance of the carbon levels through the restoration of soil organic matter. In case of sugar cane crop, the replacement of the manual cutting system, preceded by residue burning (preharvest burning), by mechanized harvest (raw sugar cane) where the straw remains on the field, favors the accumulation and differentiation of the C content in the soil. This study aimed to compare the NIRS methodology, associated with chemometrics, to the traditional reference method of dry combustion quantifying the C content in soil samples from the sugar cane agro-system. Soil samples from three situations were analyzed: burnt sugar cane, raw sugar cane and a native area, totalizing 450 samples, with levels of respectively 18.30, 22.36 and 24.72 g C kg-1. Calibration models were adjusted using PLS in the spectral data, submitted to the first derivative and smoothing with Savitz Golay with window 15. Samples were randomized and divided into 3 parts: 2/3 of the 450 samples were designated to calibration and cross validation, and the odder 1/3 to external validation. The chemometrics parameters RMSEC, RMSECV, RMSEP, R2, BIAS, CV% and RPD were chosen to estimate the quality of the model. The results of RMSEC> 1.93, RMSEP> 1.90, and R2 0.94 were satisfactory. In the native area a range of RMSEP 1.53 and R2 0.95 was obtained, while in the area of burnt sugar cane it was R2 0.50 and RMSEP 1.90. Comparing the reference technique and NIRS, the first showed a SD 0.12 and CV 0.65%, while a SD 0.18 CV 0.81% was obtained for the second, proving the efficiency of the use of NIRS to determine C content in soil

The evaluation of the soil spatial variability using a fast, robust and cheap tool is one of the key steps towards the implementation of Precision Agriculture (PA) successfully. Soil organic carbon (OC), soil total nitrogen (TN) and soil moisture content (MC) are needed to be monitored for both agriculture and environmental applications. The literature has proven that visible and near infrared (vis-NIR) spectroscopy to be a quick, cheap and robust tool to acquire information about key soil properties simultaneously with relatively high accuracy. The on-line vis-NIR measurement accuracy depends largely on the quality of calibration models. In order to establish robust calibration models for OC, TN and MC valid for few selected European farms, several factors affecting model accuracy have been studied. Nonlinear calibration techniques, e.g. artificial neural network (ANN) combined with partial least squares regression (PLSR) has provided better calibration accuracy than the linear PLSR or principal component regression analysis (PCR) alone. It was also found that effects of sample concentration statistics, including the range or standard derivation and the number of samples used for model calibration are substantial, which should be taking into account carefully. Soil MC, texture and their interaction effects are other principle factors affecting the in situ and on-line vis-NIR measurement accuracy. This study confirmed that MC is the main negative effect, whereas soil clay content plays a positive role. The general calibration models developed for soil OC, TN and MC for farms in European were validated using a previously developed vis-NIR on-line measurement system equipped with a wider vis-NIR spectrophotometer (305 – 2200 nm) than the previous version. The validation results showed this wider range on-line vis-NIR system can acquire larger than 1500 data point per ha with a very good measurement accuracy for TN and OC and excellent accuracy for MC. The validation also showed that spiking few target field samples into the general calibration models is an effective and efficient approach for upgrading the implementation of the on-line vis-NIR sensor for measurement in new fields in the selected European farms.

Submitted by Ana Hilda Fonseca (anahilda@ufba.br) on 2016-09-05T18:15:45Z No. of bitstreams: 1 Tese_Jamile_corrigida.pdf: 1767610 bytes, checksum: 59de9f8f00948e940a48efbab6f323cc (MD5) / Rejected by Vanessa Reis (vanessa.jamile@ufba.br), reason: Trabalho depositado na coleção errada trata-se de um dissertação de mestrado. Por favor efetue as devidas correções. on 2016-09-06T10:59:16Z (GMT) / Rejected by Vanessa Reis (vanessa.jamile@ufba.br), reason: Trabalho depositado na coleção errada trata-se de um dissertação de mestrado. Por favor efetue as devidas correções. on 2016-09-06T10:59:14Z (GMT) / Submitted by Ana Hilda Fonseca (anahilda@ufba.br) on 2016-09-06T13:03:08Z No. of bitstreams: 1 Tese_Jamile_corrigida.pdf: 1767610 bytes, checksum: 59de9f8f00948e940a48efbab6f323cc (MD5) / Approved for entry into archive by Uillis de Assis Santos (uillis.assis@ufba.br) on 2016-09-06T13:12:00Z (GMT) No. of bitstreams: 1 Tese_Jamile_corrigida.pdf: 1767610 bytes, checksum: 59de9f8f00948e940a48efbab6f323cc (MD5) / Made available in DSpace on 2016-09-06T13:12:00Z (GMT). No. of bitstreams: 1 Tese_Jamile_corrigida.pdf: 1767610 bytes, checksum: 59de9f8f00948e940a48efbab6f323cc (MD5) / CNPq / Este trabalho teve como objetivo desenvolver um método analítico para predição da composição química de amostras geradas no processo de produção de aromáticos utilizando a Espectroscopia no Infravermelho Próximo (NIR) associado a técnicas de calibração multivariada.Os conjuntos de calibração e validação foram selecionados através do algoritmo de Kennard-Stone e a técnica estatística utilizada para a calibração multivariada foi a Regressão por Mínimos Quadrados Parciais (PLS). As regiões espectrais selecionadas na etapa de construção dos modelos foram obtidas através do algoritmo de seleção de variáveis por regressão de mínimos quadrados parciais por intervalo (iPLS). Para escolher as condições experimentais mais adequadas para a modelagem PLS foi realizado um planejamento experimental com matriz Doehlert usando três variáveis (pré-processamento, faixa de comprimento de onda e seleção de variáveis espectrais com o algoritmo Jack-knife). Foram desenvolvidos modelos de calibrações para a previsão da concentração de não aromáticos, benzeno, tolueno, etil-benzeno, para-xileno, meta- xileno, orto-xileno, aromáticos C8s+ e aromáticos C9s+ em amostras de correntes petroquímicas; e os RMSEPs encontrados foram 0,88; 0,38; 2,43; 1,19; 1,08; 1,13; 1,29; 3,87; 1,47% (m/m), respectivamente. O desempenho do melhor modelo de calibração de cada propriedade foi avaliado por meio de parâmetros da validação externa. Com os resultados obtidos, pôde-se demonstrar que os modelos construídos foram satisfatórios e os erros encontrados são aceitáveis para controle de processo na indústria. / This work aims at developing an analytical method for predicting the chemical composition of samples generated in the production of aromatics using Near Infrared Spectroscopy (NIRS) combined with multivariate calibration techniques. The calibration and validation sets were selected by the Kennard-Stone algorithm. Partial Least Squares Regression (PLS) was the statistical technique used in the multivariate calibration. The spectral regions selected for the model development were obtained using the algorithm for spectral variable selection based on the partial least squares regression interval (iPLS). The most appropriate experimental conditions for PLS modeling were chosen using an experimental design based on a three-variable Doehlert matrix (pre-processing, wavelength range and selection of spectral variables with the Jack-knife algorithm). Calibration models were developed for predicting non-aromatic, benzene, toluene, ethyl benzene, p-xylene, m-xylene, o-xylene, aromatic C8s+ and aromatic C9s+. The performance of the best calibration model for each property was evaluated by external validation parameters. The PLS prediction of the properties presented RMSEP 0,88; 0,38; 2,43; 1,19; 1,08; 1,13; 1,29; 3,87; 1,47%w/w respectively. The PLS models showed satisfactory results and errors were found to be acceptable to the industry process control.

O desenvolvimento e aperfeiçoamento de métodos de otimização são pontos de profundo interesse em todas as áreas de pesquisa. Tais técnicas muitas vezes envolvem a aquisição de métodos de controle novos ou melhores, o que está diretamente ligado a duas tarefas importantes: a escolha de formas eficientes de monitoramento do processo e a obtenção de modelos confiáveis para a variável de interesse a partir de dados experimentais. Graças às suas diversas vantagens, os sensores óticos vêm sendo amplamente aplicados na primeira tarefa. Uma vez que é possível a utilização de vários tipos de espectroscopia através deste tipo de sensor, modelos capazes de lidar com dados espectrais estão se tornando cada vez mais atraentes. A segunda tarefa, por sua vez, depende não só de quais preditores são utilizados na construção do modelo, mas também de quantos. Como a qualidade do modelo depende também do número de variáveis selecionadas, é importante desenvolver métodos que identifiquem aqueles que explicam o máximo possível da variabilidade dos dados. O método de otimização Colônia de Formigas (ACO) aparece como uma ferramenta bastante útil na seleção de variáveis, podendo-se encontrar muitas variações desse algoritmo na literatura. O propósito deste trabalho é desenvolver métodos de seleção de variáveis com base no algoritmo ACO, conceitos estatísticos e testes de hipóteses. Para isso, diversos critérios de decisão foram implementados nas etapas do algoritmo referentes à atualização de trilha de feromônios (C1) e à seleção de modelos (C2). A fim de estudar estas modificações, foram realizados dois estudos de caso: o primeiro na área de bioprocessos e o segundo na área de caracterização de alimentos. Ambos os estudos mostraram que, em geral, os modelos com menores erros são obtidos utilizando-se métricas dos componentes do modelo, tal como o tamanho do intervalo de confiança de cada parâmetro e o teste-t de hipóteses. Além disso, a modificação do critério de seleção de modelos parece não interferir significativamente no resultado final do algoritmo. Por último, foi feito um estudo da aplicação dessas versões do ACO no campo de caracterização de combustíveis, mais especificamente diesel, associando-se duas análises espectroscópicas para predição do conteúdo de enxofre. Algumas das versões desenvolvidas mostraram-se superior ao algoritmo ACO utilizado como base para este trabalho, proposto por Ranzan (2014), e todas os versões forneceram melhores resultados na quantificação de enxofre que aqueles obtidos por PCR. Dessa forma, comprova-se a potencialidade de métricas implementadas no algoritmo ACO, associadas à espectroscopia, na seleção de preditores significativos. / The development and improvement of optimization methods are points of deep interest in all areas of research. These techniques are often related to the acquisition of new or better control methods, which are directly attached to two importante tasks: choosing efficient forms of process monitoring and obtaining reliable models for the monitored variable from experimental data. Due to their several advantagens, optical sensors are being widely applied in the first task. Since several types of spectroscopy are possible through this type of sensor, models capable of dealing with spectral data are becoming increasingly attractive. The second task depends not only on which predictors are used in the model, but also on how many. Since the quality of the model depends on the number of selected variables, it is important to develop methods that identify those that explain the greater amount of data variability as possible, without compromising the reliability of the model. The Ant Colony Optimization is an important tool for variable selection, being possible to find a lot of variations of this method in literature. The purpose of this work is to develop a method of variable selection based on the Ant Colony Optimization (ACO) algorithm, statistical concepts and hypothesis testing. For this purpose, several decision criteria for trail update (C1) and model selection (C2) were implemented within the routine. In order to study these modifications, two case study was conducted: one related to bioprocess monitoring and another one envolving the characterization of food products. Both studies showed that, in general, the models with the lowest errors were obtained through the use of model component metrics, such as the length of the confidence interval associated with each parameter and the t hypothesis test. Besides, the modification of the model selection criterion doesn’t seem to affect the algorithm final result. Finally, the aplicattion of these methods in the field of fuels characterization, specifically diesel fuel, was studied, associating two spectroscopical analyses in order to predict the sulfur content. Some of the new developed methods appeared to be better than the ACO algorithm used as basis in this work, proposed by Ranzan (2014), and all methods showed better results than those from the models constructed by PCR. Thus, it is proved the high potencial of using different metrics within ACO algorithm, associated with spectroscopy, in order to select significative predictors.

The aim of this thesis is to study and evaluate the application of hyperspectral image analysis as a non-destructive analysis method for historical mortars. This method was applied on 35 sampled mortars in varying sizes and type from the inner walls of the fortified medieval city Carcassonne. By using near infrared spectroscopy and classifying the complex multivariate data by applying the SIMCA method (Soft Independent Modelling of Class Analogies) it is possible to conduct an in depth analysis of the samples. This can then further our understanding about the construction phases as well as construction techniques used as indicated through the chemometric analysis that can identify and group the mortars in accordance to raw material and transformation process. From this could four distinct groups be found in the PCA models, two Roman periods and two high medieval periods, allowing to study Carcassonne prior to and after its enclosure. A find from the first Roman period indicates on a bathhouse or public building existing prior to the construction of the defensive wall, leading to the hypothesis that maybe more parts of the inner wall might contain older structures like this. The application of hyperspectral image analysis on historical mortars has proven itself a useful tool and simple method for studying mortars. / Målet med denna uppsats var att studera och evaluera applikationen av hyperspektral bildanalys som en icke-destruktiv analysmetod på historiskt murbruk. Instrumentet applicerades på 35 murbruksprover i varierande storlek och typ tagna från de inre murarna av den befästa medeltida staden Carcassonne. Med nära infraröd spektroskopi och klassificering av den multivariata genom SIMCA metoden (Soft Independent Modelling of Class Analogies) var det möjligt att göra en djupgående analys av proverna. Detta tillvägagångssätt kan då främja vår förståelse om stadens konstruktionsfaser och konstruktionstekniker som indikeras genom den chemometriska analysen som kan identifiera murbruket utefter råmaterial samt hur murbruket tillverkats. Från dessa metoder kunde fyra distinkta grupper finnas i PCA modellerna, två romerska perioder och två högmedeltida perioder, vilket öppnade för tolkning både innan och efter stadsmurarna rests. Ett fynd från den första romerska perioden indikerar på förekomsten av ett badhus eller publik byggnad vars väggar sedan återanvänts vid konstruktionen av den inre stadsmuren. Detta fynd leder till hypotesen att potentiellt andra delar av den inre stadsmuren kan innehålla väggar från äldre byggnader som denna. Applikationen av hyperspektral bildanalys på historiskt murbruk har påvisat sig ett användbart verktyg och simpel metod för att studera murbruk.