<p>Accommodating changing consumer tastes, nutritional targets, competitive pressures and government regulations is an ongoing task in the food industry. Product development projects tend to have competing goals and more potential solutions than can be examined efficiently. However, existing databases or spreadsheets containing formulas, ingredient properties, and product characteristics can be exploited using latent variable methods to confront difficult formulation issues. Using these methods, a product developer can target specific final product properties and systematically determine new recipes that will best meet the development objectives.</p> <p>Latent variable methods in reformulation are demonstrated for a product line of frozen muffin batters used in the food service industry. A particular attribute is to be minimized while maintaining the taste, texture, and appearance of the original products, but the minimization is difficult because the attribute in question is not well understood. Initially, existing data is used to develop a partial least squares (PLS) model, which identifies areas for further testing. Design of experiments (DOE) in the latent variable space generates new data that is used to augment the model. An optimization algorithm makes use of the updated model to produce recipes for four different products, and a significant reduction of the target attribute is achieved in all cases.</p> <p>Latent variable methods are also applied to a difficult classification problem in oat milling. Process monitoring involves manually classifying and counting the oats and hulls in the product streams of groats; a task that is time-consuming and therefore infrequent. A solution based on near infrared (NIR) imaging and PLS-discriminant analysis (PLS-DA) is investigated and found to be feasible. The PLS-DA model, built using mixed-cultivar samples, effectively separates the oats and groats into two classes. The model is validated using samples of three pure cultivars with varying moistures and growing conditions.</p> / Master of Applied Science (MASc)

<p>This thesis provides new regression methods for the removal of structured noise in datasets. With multivariable data, the variables and the noise can be both temporally correlated (i.e. auto correlated in time) and contemporaneously correlated (i.e. cross-correlated at the same time). In many occasions it is possible to acquire measurements of the noise, or some function of it, during the data collection. Several new constrained latent variable methods (LVM) that are built upon previous LVM regression frameworks are introduced. These methods make use of the additional information available about the noise to decompose a dataset into basis for the noise and signal. The properties of these methods are investigated mathematically, and through both simulation and application to actual biomedical data.</p> <p>In Chapter Two, linear, constrained LVM methods are introduced. The performance of these methods are compared to the other similar LVM methods as well as ordinary PLS throughout several simulation studies. In Chapter Three, a NIPALS type algorithm is developed for the soft constrained PLS method which is also able to account for missing data as well as datasets with large covariance matrices. Chapter Four introduces the nonlinear-kernelized constrained LVM methods. These methods are capable of handling severe nonlinearities in the datasets. The performance of these methods are compared to nonlinear kernel PLS method. In Chapter Five the constrained methods are used to remove ballistocardiographic and muscle artifacts from EEG datasets in combined EEG-fMRI as well as single EEG experiments on patients. The results are shown and compared to the standard noise removal methods used in the field. Finally in Chapter Six, the overall conclusion and scope of the future work is laid out.</p> / Doctor of Philosophy (PhD)

L’objectif de cette thèse était d’appliquer des méthodes d’analyse multivariées au traitement des données provenant de la spectroscopie de plasma induit par laser (LIBS) et de la spectroscopie térahertz (THz) dans le but d’accroître les performances analytiques de ces techniques.Les spectres LIBS provenaient de campagnes de mesures directes sur différents sites géologiques. Une approche univariée n’a pas été envisageable à cause d’importants effets de matrices et c’est pour cela qu’on a analysé les données provenant des spectres LIBS par réseaux de neurones artificiels (ANN). Cela a permis de quantifier plusieurs éléments mineurs et majeurs dans les échantillons de sol avec un écart relatif de prédiction inférieur à 20% par rapport aux valeurs de référence, jugé acceptable pour des analyses sur site. Dans certains cas, il a cependant été nécessaire de prendre en compte plusieurs modèles ANN, d’une part pour classer les échantillons de sol en fonction d’un seuil de concentration et de la nature de leur matrice, et d’autre part pour prédire la concentration d’un analyte. Cette approche globale a été démontrée avec succès dans le cas particulier de l’analyse du plomb pour un échantillon de sol inconnu. Enfin, le développement d’un outil de traitement par ANN a fait l’objet d’un transfert industriel.Dans un second temps, nous avons traité des spectres d’absorbance terahertz. Ce spectres provenaient de mesures d’absorbance sur des mélanges ternaires de Fructose-Lactose-acide citrique liés par du polyéthylène et préparés sous forme de pastilles. Une analyse semi-quantitative a été réalisée avec succès par analyse en composantes principales (ACP). Puis les méthodes quantitatives de régression par moindres carrés partiels (PLS) et de réseaux de neurons artificiels (ANN) ont permis de prédire les concentrations de chaque constituant de l’échantillon avec une valeur d’erreur quadratique moyenne inférieure à 0.95 %. Pour chaque méthode de traitement, le choix des données d’entrée et la validation de la méthode ont été discutés en détail. / The aim of this work was the application of multivariate methods to analyze spectral data from laser-induced breakdown spectroscopy (LIBS) and terahertz (THz) spectroscopy to improve the analytical ability of these techniques.In this work, the LIBS data were derived from on-site measurements of soil samples. The common univariate approach was not efficient enough for accurate quantitative analysis and consequently artificial neural networks (ANN) were applied. This allowed quantifying several major and minor elements into soil samples with relative error of prediction lower than 20% compared to reference values. In specific cases, a single ANN model didn’t allow to successfully achieving the quantitative analysis and it was necessary to exploit a series of ANN models, either for classification purpose against a concentration threshold or a matrix type, or for quantification. This complete approach based on a series of ANN models was efficiently applied to the quantitative analysis of unknown soil samples. Based on this work, a module of data treatment by ANN was included into the software Analibs of the IVEA company. The second part of this work was focused on the data treatment of absorbance spectra in the terahertz range. The samples were pressed pellets of mixtures of three products, namely fructose, lactose and citric acid with polyethylene as binder. A very efficient semi-quantitative analysis was conducted by using principal component analysis (PCA). Then, quantitative analyses based on partial least squares regression (PLS) and ANN allowed quantifying the concentrations of each product with a root mean square error (RMSE) lower than 0.95 %. All along this work on data processing, both the selection of input data and the evaluation of each model have been studied in details.

Para um composto químico exercer seu efeito bioativo é necessário que ele atravesse várias barreiras biológicas até alcançar seu sitio de ação. Propriedades farmacocinéticas insatisfatórias (como absorção, distribuição, metabolismo e excreção) são reconhecidamente as principais causas na descontinuidade de pesquisas na busca por novos fármacos. Neste trabalho, modelos biofísicos foram utilizados para o estudo de absorção de uma série de flavonóides naturais com atividade tripanossomicida. O coeficiente cromatográfico de partição, kw, foi determinado através da cromatografia líquida de alta eficiência em fase reversa, RP-HPLC, utilizando-se de colunas cromatográficas empacotadas com constituintes básicos da membrana biológica (fosfatidilcolina e colesterol). Os resultados obtidos demonstraram que nas colunas compostas por fosfatidilcolina a retenção de flavonóides hidroxilados é determinada por interações secundárias, além da partição, e no caso da coluna de colesterol, a partição é o principal mecanismo que rege a retenção. Uma série de descritores físico-químicos foi gerada pelos campos moleculares de interações (MIFs) entre os flavonóides naturais e algumas sondas químicas virtuais, utilizando o programa GRID. Os descritores físico-químicos gerados foram correlacionados com os log kw por análise dos mínimos múltiplos parciais (PLS), utilizando o programa VolSurf, com a finalidade de gerar um modelo quantitativo entre estrutura e propriedade (QSPR) para esta classe de compostos. O modelo produzido por este estudo, ao utilizar os dados de partição em colesterol, log kwCol, apresentou elevada consistência interna, com bom poder de correlação (R2 = 0, 97) e predição (Q2 = 0,86) para a partição destas moléculas / In order to a chemical compound exert its bioactive effect it is necessary that it crosses some biological barriers until reaching its site of action. Unfavorable pharmacokinetics properties (absorption, distribution, metabolism and excretion) are admittedly one of the main causes in the discontinuity of research in the search for new drugs. In this work, biophysics models were used for the study of absorption of a series of natural flavonoids with trypanocide activity. The chromatographic retention indices (log kw) were determined on immobilized artificial membranes columns (IAM.PC.DD, IAM.PC.DD2, Cholesteryl 10-Undecetonoato) obtained by the extrapolation method. The results demonstrated that in the composed columns for fosfatidilcolina the retention of hydroxil flavonoids is determined by secondary interactions, beyond the partition. In the case of the retention for the cholesterol column, the partition is the main mechanism that drives the retention. A series of physico-chemical descriptors were generated by the molecular interaction fields (MIF) between the flavonoids and some virtual chemical probes, using the program GRID. The descriptors were correlated with log kw by the partial least squares regression (PLS), using the VolSurf program, with the purpose to generate a quantitative model between the structure and the retention (QSRR) for this compounds class. The model produced for this study, when using the data of partition in cholesterol, log kwCol, presented high internal consistency, with good correlation power (R2 = 0, 97) and prediction (Q2 = 0,86) for the partition of these molecules

Para um composto químico exercer seu efeito bioativo é necessário que ele atravesse várias barreiras biológicas até alcançar seu sitio de ação. Propriedades farmacocinéticas insatisfatórias (como absorção, distribuição, metabolismo e excreção) são reconhecidamente as principais causas na descontinuidade de pesquisas na busca por novos fármacos. Neste trabalho, modelos biofísicos foram utilizados para o estudo de absorção de uma série de flavonóides naturais com atividade tripanossomicida. O coeficiente cromatográfico de partição, kw, foi determinado através da cromatografia líquida de alta eficiência em fase reversa, RP-HPLC, utilizando-se de colunas cromatográficas empacotadas com constituintes básicos da membrana biológica (fosfatidilcolina e colesterol). Os resultados obtidos demonstraram que nas colunas compostas por fosfatidilcolina a retenção de flavonóides hidroxilados é determinada por interações secundárias, além da partição, e no caso da coluna de colesterol, a partição é o principal mecanismo que rege a retenção. Uma série de descritores físico-químicos foi gerada pelos campos moleculares de interações (MIFs) entre os flavonóides naturais e algumas sondas químicas virtuais, utilizando o programa GRID. Os descritores físico-químicos gerados foram correlacionados com os log kw por análise dos mínimos múltiplos parciais (PLS), utilizando o programa VolSurf, com a finalidade de gerar um modelo quantitativo entre estrutura e propriedade (QSPR) para esta classe de compostos. O modelo produzido por este estudo, ao utilizar os dados de partição em colesterol, log kwCol, apresentou elevada consistência interna, com bom poder de correlação (R2 = 0, 97) e predição (Q2 = 0,86) para a partição destas moléculas / In order to a chemical compound exert its bioactive effect it is necessary that it crosses some biological barriers until reaching its site of action. Unfavorable pharmacokinetics properties (absorption, distribution, metabolism and excretion) are admittedly one of the main causes in the discontinuity of research in the search for new drugs. In this work, biophysics models were used for the study of absorption of a series of natural flavonoids with trypanocide activity. The chromatographic retention indices (log kw) were determined on immobilized artificial membranes columns (IAM.PC.DD, IAM.PC.DD2, Cholesteryl 10-Undecetonoato) obtained by the extrapolation method. The results demonstrated that in the composed columns for fosfatidilcolina the retention of hydroxil flavonoids is determined by secondary interactions, beyond the partition. In the case of the retention for the cholesterol column, the partition is the main mechanism that drives the retention. A series of physico-chemical descriptors were generated by the molecular interaction fields (MIF) between the flavonoids and some virtual chemical probes, using the program GRID. The descriptors were correlated with log kw by the partial least squares regression (PLS), using the VolSurf program, with the purpose to generate a quantitative model between the structure and the retention (QSRR) for this compounds class. The model produced for this study, when using the data of partition in cholesterol, log kwCol, presented high internal consistency, with good correlation power (R2 = 0, 97) and prediction (Q2 = 0,86) for the partition of these molecules

Das Messverfahren der optischen Spektrophotometrie wird zur kontinuierlichen Messung der Abwasserqualität auf ihre Einsatztauglichkeit überprüft. Der chemische Sauerstoffbedarf (CSB) wird als zentraler Kennwert für die stoffliche Verschmutzung von Abwasser und für dessen Nachweis in Oberflächengewässern eingesetzt, welche es zu bestimmen galt. Dabei wird der Informationsgehalt über eine organische, summarische Kohlenstoffbelastung mittels einer zusätzlichen Fraktionierung erhöht. In einer Labormesskampagne werden auf der Grundlage von Respirationsversuchen Daten aus Extinktionswerten des UV/Vis-Spektrums und Referenzwerten (Standardanalyseparameter und simulierte Stoffkonzentrationen mithilfe des Activated Sludge Modell No. 1) generiert. Darauf aufbauend werden Kalibrationsmodelle für den CSB und einzelne Fraktionen entwickelt. Die Modelle werden mithilfe des Regressionsansatzes der Partial-Least-Squares entwickelt und im Rahmen eines Anwendungsbeispiels auf ihre Praxistauglichkeit überprüft. Als Ergebnis dieser Arbeit stehen Kalibrationsmodelle für den Einsatz im kommunalem Abwasser unter Trockenwetterbedingungen zur Verfügung. Die Vorhersagequalität nimmt mit zunehmender Differenzierung ab. Von einer Weiterverwendung der berechneten Äquivalentkonzentrationen für die CSB-Fraktionen (SS, XS, SI und XI), z. B. als Kalibriergröße für Stofftransportmodelle oder als Steuer- und Regelgröße, wird allerdings abgeraten. Als Ursache für die hohen Messungenauigkeiten wurde eine unzureichende Anpassung an die Veränderungen in der Abwasserzusammensetzung während eines Trockenwettertagesganges identifiziert. Mit einer erweiterten Datengrundlage, unter der Verwendung von Standardanalyseparametern (CSB, CSBmf und BSB) in einer Abwasserprobe, welche für den Ausschluss von Stoffverbindungen vor und nach einer respirativen Vorbehandlung bestimmt werden, wird eine höhere Modellgüte in Aussicht gestellt. Darüber hinaus wird ein Umdenken hinsichtlich statischer - hin zu dynamischen - Kalibrationsfunktionen für UV/Vis-Sensoren vorgeschlagen. Eine Generalisierbarkeit der entwickelten Kalibrationsmodelle auf weitere Wetterbedingungen, Messstandorte oder Sensoren wird nicht empfohlen.:Abbildungen VI Tabellen XIII Abkürzungen XV 1 Einleitung 1 1.1 Motivation 1 1.2 Zielstellung 2 2 Stand der Forschung 5 2.1 Kohlenstoffe 6 2.1.1 Zusammensetzung und Herkunft im häuslichen Abwasser 7 2.1.1.1 Fette 8 2.1.1.2 Proteine 8 2.1.1.3 Tenside 9 2.1.1.4 Phenole 10 2.1.1.5 Kohlenwasserstoffe 10 2.1.2 Fraktionierung von Kohlenstoffverbindungen 11 2.1.2.1 Chemischer Sauerstoffbedarf 12 2.1.2.2 Ansätze zur CSB-Fraktionierung 12 2.1.2.3 Stoffzusammensetzung einzelner CSB-Fraktionen 15 2.1.2.4 Messmethoden zur Bestimmung des CSB 18 2.2 Optische Spektroskopie 20 2.2.1 Grundlagen 20 2.2.1.1 Elektromagnetische Strahlung 20 2.2.1.2 Einordnung der optischen Spektroskopie 21 2.2.1.3 Lichtabsorption 21 2.2.1.4 Chemisch-physikalische Grundlagen 22 2.2.1.5 Mathematische Grundlagen 24 2.2.1.6 Extinktionsmessung 25 2.2.2 Online-Messtechnik 26 2.2.2.1 Sensoren /-hersteller 26 2.2.2.2 Kalibrierung 26 2.2.2.2.1 Kalibrierung der S::CAN MESSTECHNIK GmbH 27 2.2.2.2.2 Unabhängige Analyseverfahren zur Auswertung spektrophotometrischer Messreihen 28 2.2.2.3 Messung 29 2.2.2.3.1 Einstellungen und Voraussetzungen 29 2.2.2.3.2 Qualitative Einflussnahme von Störgrößen auf die spektroskopische Datenerfassung 30 2.2.3 Einsatz in der Siedlungswasserwirtschaft und Hydrologie 31 3 Versuchsdurchführung und Analytik 33 3.1 Messkampagnen 33 3.1.1 Labormessversuche 33 3.1.1.1 Respirationsversuch 34 3.1.1.1.1 Versuchsaufbau zum Respirationsversuch 35 3.1.1.1.2 Betriebshinweise Respirationsversuch 38 3.1.1.2 Verdünnungsversuch 41 3.1.1.2.1 Versuchsaufbau zum Verdünnungsversuch 42 3.1.1.2.2 Betriebshinweise Verdünnungsversuch 43 3.1.2 Feldmessversuch 43 3.1.2.1 Versuchsaufbau zum Feldmessversuch 44 3.1.2.2 Betriebshinweise Feldmessversuch 46 3.2 Abwasserproben: Aufbewahrung und Analytik 47 3.2.1 Konservierung und Probenvorbehandlung 48 3.2.2 Standardisierte Laboranalyseverfahren 49 3.2.2.1 CSB 49 3.2.2.2 Biologischer Sauerstoffbedarf BSBn 50 3.3 Mess- und Regelinstrumente 51 3.3.1 Optischer Multiparameter-Sensor 51 3.3.2 Luminescent Dissolved Oxygen-Sensor (LDO) 53 3.3.3 Peristaltik-Pumpe 54 3.3.4 Dispergierer 54 4 Untersuchungen zur Entwicklung und Anwendung von UV/Vis-Kalibrierungen 55 4.1 Statistische Verfahren zur Kalibrierung 55 4.1.1 Datengrundlage und Methoden 56 4.1.1.1 Datengrundlage 56 4.1.1.2 Multivariate Datenanalyse 57 4.1.1.2.1 Regressionsanalyse 58 4.1.1.2.1.1 Schätzung der Regressionsfunktion 59 4.1.1.2.2 Qualitätsprüfung 61 4.1.1.2.3 Prüfung der Modellprämissen 63 4.1.1.2.4 Multivariate Regressionsanalyse 66 4.1.1.3 Vergleich der Kalibrierverfahren 70 4.1.2 Ergebnisse 70 4.1.2.1 Regressionsansätze für UV/Vis-Kalibrierung 70 4.1.2.1.1 Partial-Least-Squares Regression (PLS-R) 70 4.1.2.1.2 Lasso-Regression 73 4.1.2.1.3 Herstellerkalibrierung (SCAN GmbH) 73 4.1.2.1.3.1 Anwendung der globalen Herstellerkalibrierung 73 4.1.2.1.3.2 Lokal angepasste Herstellerkalibrierung 74 4.1.3 Auswertungen 75 4.1.3.1 Tauglichkeit angewandter Regressionsansätze zur Entwicklung von UV/Vis-Kalibrierfunktionen 75 4.1.3.1.1 Vergleich der Vorhersagequalitäten zwischen Regressionsansätzen und Herstellerkali¬- brierung 75 4.1.3.1.2 Aussagekraft angewandter Regressionsmodelle 77 4.1.3.1.2.1 Regressionsfunktion und -koeffizienten 77 4.1.3.1.2.2 Modellprämissen 78 4.1.3.2 Identifizierung signifikanter WL oder -Bereiche 80 4.2 Fraktionierung von CSB-Verbindungen 81 4.2.1 Datengrundlage und Methoden 82 4.2.1.1 Laborwertmethode 83 4.2.1.2 Modellwertmethode 85 4.2.1.2.1 Respirometrische Messung 86 4.2.1.2.2 Sauerstoffverbrauchsrate 87 4.2.1.2.3 Modellberechnung 89 4.2.1.2.4 Simulationsmethode mit modifiziertem Activated Sludge Modell No. 1 92 4.2.1.2.5 Modellkalibrierung 95 4.2.1.2.6 Datenauswahl 96 4.2.1.3 Lichtabsorptionsmethode 96 4.2.2 Ergebnisse 97 4.2.2.1 Modellwertmethode mit ASM No. 1 97 4.2.2.2 Auswahl von Modelldaten 100 4.2.2.3 UV/Vis-Kalibrierfunktionen 101 4.2.2.3.1 CSB-Fraktionen 101 4.2.2.3.2 Vergleich MW- und LW-Modell 103 4.2.3 Auswertungen 104 4.2.3.1 Tauglichkeit von Simulationsergebnissen aus Modellwertmethode zur Entwicklung von Kalibrierfunktionen 104 4.2.3.2 Abweichende Vorhersagequalitäten zwischen den UV/Vis-Kalibrierfunktionen 106 4.2.3.3 Messunsicherheiten und Modellqualität 107 4.2.3.4 Signifikante Wellenlängen oder -bereiche für einzelne CSB-Fraktionen 109 4.3 Anwendungsbeispiel: Kohlenstoffumsatz entlang einer Fließstrecke 111 4.3.1 Datengrundlage und Methoden 112 4.3.1.1 Einsatz von UV/Vis-Messtechnik 115 4.3.1.1.1 Vergleichbarkeit bei Parallelbetrieb baugleicher Sensoren 115 4.3.1.1.1.1 Versuchsdurchführung 116 4.3.1.1.1.2 Berechnungsansätze 116 4.3.1.1.2 Lokale Kalibrierung 117 4.3.1.1.2.1 Univariat 118 4.3.1.1.2.2 Multivariat 118 4.3.1.2 Kohlenstoffumwandlung und -umsatz innerhalb des Durchflussreaktors 118 4.3.1.2.1 Vorverarbeitung von UV/Vis-Daten 120 4.3.1.2.2 Zeitsynchronisation mithilfe der Fließzeit 120 4.3.1.2.3 Bestimmung von stofflichen Veränderungen in einem Wasserpaket 121 4.3.2 Ergebnisse 122 4.3.2.1 Praxiseinsatz von UV/Vis-Messtechnik 122 4.3.2.1.1 Stabilität und Vergleichbarkeit von Messsignalen bei unterschiedlichen Sensoren 122 4.3.2.1.1.1 Messgüte 122 4.3.2.1.1.2 Sensoranpassung 124 4.3.2.1.2 UV/Vis-Kalibrationsfunktionen 125 4.3.2.1.2.1 Validierung LK-PLS-R 126 4.3.2.1.2.2 Lokale Nachkalibrierung LK-PLS-R 128 4.3.2.1.3 Anwendung entwickelter Kalibrationsmodelle auf Zeitreihen 130 4.3.2.2 Kohlenstoffumsatz 131 4.3.3 Auswertungen 135 4.3.3.1 Tauglichkeit von UV/Vis-Messtechnik für den Einsatz in der Kanalisation 135 4.3.3.1.1 Vorhersagegenauigkeit von Kalibrationsfunktionen 135 4.3.3.1.2 Abweichende Messergebnisse der Extinktion von einzelnen Sensoren 135 4.3.3.2 Veränderungen in den Konzentrationen einzelner Kohlenstofffraktionen entlang der Fließstrecke 136 5 Diskussion 139 6 Ausblick 151 7 Zusammenfassung 153 8 Literaturverzeichnis 157 A Anhang 171 A.1 Respirationsversuche CSB-Fraktionen 171 A.1.1 Quellcode - CSB-Fraktionierung 171 A.1.2 Respirationsversuche CSB-Fraktionen 175 A.1.3 Quellcode - PLS-Regression 178 A.1.4 UV/Vis-Kalibrierung - CSB-Fraktionen 180 A.1.5 Modellgüte 183 A.1.6 Modellprämissen 184 A.2 Feldmesskampagne 188 A.2.1 Sensorkompensation 188 A.2.2 Korrelationsplots 189 A.2.2.1 Validierung der Kalibrationsmodelle 189 A.2.2.2 Nachkalibrierung der Kalibrationsmodelle 192 A.2.2.2.1 univariat 192 A.2.3 Stoffliche Veränderungen in Wasserpaketen 198 A.2.4 Laboranalysen Stoffliche Veränderungen in Wasserpaketen 201 / Optical spectrophotometry is checked as measuring method for continuous monitoring of waste water quality. The chemical oxygen demand (COD) is used as a central parame-ter for the material assessment of waste water and for its detection in surface waters. The information value about an organic load is increased using an additional fractiona-tion. In a laboratory measurement campaign, data from extinction values of the UV/Vis spectrum and reference values are created (standard analysis parameters and simulated concentrations by using the Activated Sludge Model No. 1). Based on this calibration models for the COD and individual fractions are developed using the regression ap-proach of the partial least squares and their practical suitability is checked in the context of an application example. As a result of this work, calibration models for use in munici-pal wastewater under dry weather conditions, are available. The prediction quality de-creases with increasing differentiation. We advise against further use of the calculated equivalent concentrations for the COD fractions (SS, XS, SI und XI), e.g. as a calibration var-iable for mass transfer models or as a control and regulation variable. The reason for higher measurement uncertainties is identified as insufficient adaptation to the changing wastewater composition during a dry weather day. With an extended data basis, a higher model quality is promised: Standard analysis parameters (COD, CODmf and BOD) are de-termined in wastewater samples before and after respiratory pretreatment in order to be able to rule out substances. In addition, rethinking of static calibration functions for UV/Vis sensors is proposed towards dynamic methods. A generalization of calibration models to other weather conditions, measurement locations or sensors is not recom-mended.:Abbildungen VI Tabellen XIII Abkürzungen XV 1 Einleitung 1 1.1 Motivation 1 1.2 Zielstellung 2 2 Stand der Forschung 5 2.1 Kohlenstoffe 6 2.1.1 Zusammensetzung und Herkunft im häuslichen Abwasser 7 2.1.1.1 Fette 8 2.1.1.2 Proteine 8 2.1.1.3 Tenside 9 2.1.1.4 Phenole 10 2.1.1.5 Kohlenwasserstoffe 10 2.1.2 Fraktionierung von Kohlenstoffverbindungen 11 2.1.2.1 Chemischer Sauerstoffbedarf 12 2.1.2.2 Ansätze zur CSB-Fraktionierung 12 2.1.2.3 Stoffzusammensetzung einzelner CSB-Fraktionen 15 2.1.2.4 Messmethoden zur Bestimmung des CSB 18 2.2 Optische Spektroskopie 20 2.2.1 Grundlagen 20 2.2.1.1 Elektromagnetische Strahlung 20 2.2.1.2 Einordnung der optischen Spektroskopie 21 2.2.1.3 Lichtabsorption 21 2.2.1.4 Chemisch-physikalische Grundlagen 22 2.2.1.5 Mathematische Grundlagen 24 2.2.1.6 Extinktionsmessung 25 2.2.2 Online-Messtechnik 26 2.2.2.1 Sensoren /-hersteller 26 2.2.2.2 Kalibrierung 26 2.2.2.2.1 Kalibrierung der S::CAN MESSTECHNIK GmbH 27 2.2.2.2.2 Unabhängige Analyseverfahren zur Auswertung spektrophotometrischer Messreihen 28 2.2.2.3 Messung 29 2.2.2.3.1 Einstellungen und Voraussetzungen 29 2.2.2.3.2 Qualitative Einflussnahme von Störgrößen auf die spektroskopische Datenerfassung 30 2.2.3 Einsatz in der Siedlungswasserwirtschaft und Hydrologie 31 3 Versuchsdurchführung und Analytik 33 3.1 Messkampagnen 33 3.1.1 Labormessversuche 33 3.1.1.1 Respirationsversuch 34 3.1.1.1.1 Versuchsaufbau zum Respirationsversuch 35 3.1.1.1.2 Betriebshinweise Respirationsversuch 38 3.1.1.2 Verdünnungsversuch 41 3.1.1.2.1 Versuchsaufbau zum Verdünnungsversuch 42 3.1.1.2.2 Betriebshinweise Verdünnungsversuch 43 3.1.2 Feldmessversuch 43 3.1.2.1 Versuchsaufbau zum Feldmessversuch 44 3.1.2.2 Betriebshinweise Feldmessversuch 46 3.2 Abwasserproben: Aufbewahrung und Analytik 47 3.2.1 Konservierung und Probenvorbehandlung 48 3.2.2 Standardisierte Laboranalyseverfahren 49 3.2.2.1 CSB 49 3.2.2.2 Biologischer Sauerstoffbedarf BSBn 50 3.3 Mess- und Regelinstrumente 51 3.3.1 Optischer Multiparameter-Sensor 51 3.3.2 Luminescent Dissolved Oxygen-Sensor (LDO) 53 3.3.3 Peristaltik-Pumpe 54 3.3.4 Dispergierer 54 4 Untersuchungen zur Entwicklung und Anwendung von UV/Vis-Kalibrierungen 55 4.1 Statistische Verfahren zur Kalibrierung 55 4.1.1 Datengrundlage und Methoden 56 4.1.1.1 Datengrundlage 56 4.1.1.2 Multivariate Datenanalyse 57 4.1.1.2.1 Regressionsanalyse 58 4.1.1.2.1.1 Schätzung der Regressionsfunktion 59 4.1.1.2.2 Qualitätsprüfung 61 4.1.1.2.3 Prüfung der Modellprämissen 63 4.1.1.2.4 Multivariate Regressionsanalyse 66 4.1.1.3 Vergleich der Kalibrierverfahren 70 4.1.2 Ergebnisse 70 4.1.2.1 Regressionsansätze für UV/Vis-Kalibrierung 70 4.1.2.1.1 Partial-Least-Squares Regression (PLS-R) 70 4.1.2.1.2 Lasso-Regression 73 4.1.2.1.3 Herstellerkalibrierung (SCAN GmbH) 73 4.1.2.1.3.1 Anwendung der globalen Herstellerkalibrierung 73 4.1.2.1.3.2 Lokal angepasste Herstellerkalibrierung 74 4.1.3 Auswertungen 75 4.1.3.1 Tauglichkeit angewandter Regressionsansätze zur Entwicklung von UV/Vis-Kalibrierfunktionen 75 4.1.3.1.1 Vergleich der Vorhersagequalitäten zwischen Regressionsansätzen und Herstellerkali¬- brierung 75 4.1.3.1.2 Aussagekraft angewandter Regressionsmodelle 77 4.1.3.1.2.1 Regressionsfunktion und -koeffizienten 77 4.1.3.1.2.2 Modellprämissen 78 4.1.3.2 Identifizierung signifikanter WL oder -Bereiche 80 4.2 Fraktionierung von CSB-Verbindungen 81 4.2.1 Datengrundlage und Methoden 82 4.2.1.1 Laborwertmethode 83 4.2.1.2 Modellwertmethode 85 4.2.1.2.1 Respirometrische Messung 86 4.2.1.2.2 Sauerstoffverbrauchsrate 87 4.2.1.2.3 Modellberechnung 89 4.2.1.2.4 Simulationsmethode mit modifiziertem Activated Sludge Modell No. 1 92 4.2.1.2.5 Modellkalibrierung 95 4.2.1.2.6 Datenauswahl 96 4.2.1.3 Lichtabsorptionsmethode 96 4.2.2 Ergebnisse 97 4.2.2.1 Modellwertmethode mit ASM No. 1 97 4.2.2.2 Auswahl von Modelldaten 100 4.2.2.3 UV/Vis-Kalibrierfunktionen 101 4.2.2.3.1 CSB-Fraktionen 101 4.2.2.3.2 Vergleich MW- und LW-Modell 103 4.2.3 Auswertungen 104 4.2.3.1 Tauglichkeit von Simulationsergebnissen aus Modellwertmethode zur Entwicklung von Kalibrierfunktionen 104 4.2.3.2 Abweichende Vorhersagequalitäten zwischen den UV/Vis-Kalibrierfunktionen 106 4.2.3.3 Messunsicherheiten und Modellqualität 107 4.2.3.4 Signifikante Wellenlängen oder -bereiche für einzelne CSB-Fraktionen 109 4.3 Anwendungsbeispiel: Kohlenstoffumsatz entlang einer Fließstrecke 111 4.3.1 Datengrundlage und Methoden 112 4.3.1.1 Einsatz von UV/Vis-Messtechnik 115 4.3.1.1.1 Vergleichbarkeit bei Parallelbetrieb baugleicher Sensoren 115 4.3.1.1.1.1 Versuchsdurchführung 116 4.3.1.1.1.2 Berechnungsansätze 116 4.3.1.1.2 Lokale Kalibrierung 117 4.3.1.1.2.1 Univariat 118 4.3.1.1.2.2 Multivariat 118 4.3.1.2 Kohlenstoffumwandlung und -umsatz innerhalb des Durchflussreaktors 118 4.3.1.2.1 Vorverarbeitung von UV/Vis-Daten 120 4.3.1.2.2 Zeitsynchronisation mithilfe der Fließzeit 120 4.3.1.2.3 Bestimmung von stofflichen Veränderungen in einem Wasserpaket 121 4.3.2 Ergebnisse 122 4.3.2.1 Praxiseinsatz von UV/Vis-Messtechnik 122 4.3.2.1.1 Stabilität und Vergleichbarkeit von Messsignalen bei unterschiedlichen Sensoren 122 4.3.2.1.1.1 Messgüte 122 4.3.2.1.1.2 Sensoranpassung 124 4.3.2.1.2 UV/Vis-Kalibrationsfunktionen 125 4.3.2.1.2.1 Validierung LK-PLS-R 126 4.3.2.1.2.2 Lokale Nachkalibrierung LK-PLS-R 128 4.3.2.1.3 Anwendung entwickelter Kalibrationsmodelle auf Zeitreihen 130 4.3.2.2 Kohlenstoffumsatz 131 4.3.3 Auswertungen 135 4.3.3.1 Tauglichkeit von UV/Vis-Messtechnik für den Einsatz in der Kanalisation 135 4.3.3.1.1 Vorhersagegenauigkeit von Kalibrationsfunktionen 135 4.3.3.1.2 Abweichende Messergebnisse der Extinktion von einzelnen Sensoren 135 4.3.3.2 Veränderungen in den Konzentrationen einzelner Kohlenstofffraktionen entlang der Fließstrecke 136 5 Diskussion 139 6 Ausblick 151 7 Zusammenfassung 153 8 Literaturverzeichnis 157 A Anhang 171 A.1 Respirationsversuche CSB-Fraktionen 171 A.1.1 Quellcode - CSB-Fraktionierung 171 A.1.2 Respirationsversuche CSB-Fraktionen 175 A.1.3 Quellcode - PLS-Regression 178 A.1.4 UV/Vis-Kalibrierung - CSB-Fraktionen 180 A.1.5 Modellgüte 183 A.1.6 Modellprämissen 184 A.2 Feldmesskampagne 188 A.2.1 Sensorkompensation 188 A.2.2 Korrelationsplots 189 A.2.2.1 Validierung der Kalibrationsmodelle 189 A.2.2.2 Nachkalibrierung der Kalibrationsmodelle 192 A.2.2.2.1 univariat 192 A.2.3 Stoffliche Veränderungen in Wasserpaketen 198 A.2.4 Laboranalysen Stoffliche Veränderungen in Wasserpaketen 201

[ES] La presente tesis doctoral está dedicada a estudiar, desarrollar y aplicar metodologías basadas en datos, fundamentadas en modelos estadísticos multivariantes de variables latentes, para abordar el paradigma del aprendizaje científico en el entorno de la Industria 4.0. Se pone especial énfasis en los modelos causales basados en variables latentes que utilizan tanto datos provenientes de un diseño de experimentos como, principalmente, datos provenientes del proceso de producción diario, es decir, datos históricos. La tesis está estructurada en cinco partes. La primera parte discute el paradigma del aprendizaje científico en el entorno de la Industria 4.0. Se destacan los objetivos de la tesis. Además, se presenta una descripción exhaustiva de los modelos basados en variables latentes, sobre los cuales se fundamentan las metodologías novedosas propuestas en esta tesis. En la segunda parte, se presentan las novedosas aportaciones metodológicas. En primer lugar, se muestra el potencial de PLS para analizar datos del DOE, con o sin datos faltantes. Posteriormente, el potencial de los modelos causales basados en variables latentes se centra en definir el espacio de diseño de la materia prima que proporciona garantía de calidad con un cierto nivel de confianza para los atributos críticos de calidad, junto con el desarrollo de un nuevo índice de capacidad multivariante basado en el espacio latente para clasificar y seleccionar proveedores para una materia prima particular utilizada en un proceso de fabricación. La tercera parte pretende abordar aplicaciones novedosas mediante modelos causales basados en variables latentes utilizando datos históricos. En primer lugar, se trata de su aplicación en el ámbito sanitario: la Pandemia COVID-19. En este contexto, se utiliza el uso de modelos basados en variables latentes para desarrollar una alternativa a los ensayos clínicos controlados con placebo. Luego, se utilizan modelos basados en variables latentes para optimizar procesos en el marco de aplicaciones industriales. La cuarta parte presenta una interfaz gráfica de usuario desarrollada en código Python que integra los métodos desarrollados con el objetivo de ser autoexplicativa y fácil de usar. Finalmente, la última parte discute la relevancia de esta disertación, incluyendo propuestas que merecen mayor investigación. / [CA] Aquesta tesi doctoral està dedicada a estudiar, desenvolupar i aplicar metodologies basades en dades, fonamentades en models estadístics multivariants de variables latents, per abordar el paradigma de l'aprenentatge científic a l'entorn de la Indústria 4.0. Es posa un èmfasi especial en els models causals basats en variables latents que utilitzen tant; dades provinents d'un disseny d'experiments com, principalment, dades provinents del procés de producció diari, és a dir, dades històriques. La tesi està estructurada en cinc parts. A la primera part es discuteix el paradigma de l'aprenentatge científic a l'entorn de la Indústria 4.0. Es destaquen els objectius de la tesi. A més, es presenta una descripció exhaustiva dels models basats en variables latents, sobre els quals es fonamenten les noves metodologies proposades en aquesta tesi. A la segona part, es presenten les noves aportacions metodològiques. En primer lloc, es mostra el potencial de PLS per analitzar dades del DOE, amb dades faltants o sense aquestes. Posteriorment, el potencial dels models causals basats en variables latents se centra a definir l'espai de disseny de la matèria prima que proporciona garantia de qualitat amb un cert nivell de confiança per als atributs crítics de qualitat, juntament amb el desenvolupament d'un nou índex de capacitat multivariant basat en l'espai latent per a classificar i seleccionar proveïdors per a una primera matèria particular utilitzada en un procés de fabricació. La tercera part pretén abordar aplicacions noves mitjançant models causals basats en variables latents utilitzant dades històrques. En primer lloc, es tracta de la seva aplicació a l'àmbit sanitari: la Pandèmia COVID-19. En aquest context, es fa servir l'ús de models basats en variables latents per desenvolupar una alternativa als assaigs clínics controlats amb placebo. Després s'utilitzen models basats en variables latents per optimitzar processos en el marc d'aplicacions industrials. La quarta part presenta una interfície gràfica d'usuari desenvolupada en codi Python que integra els mètodes desenvolupats amb l'objectiu de ser autoexplicativa i fàcil d'usar. Finalment, l'última part discuteix la rellevància d'aquesta dissertació, incloent-hi propostes que mereixen més investigació. / [EN] The present Ph.D. thesis is devoted to studying, developing, and applying data-driven methodologies, based on multivariate statistical models of latent variables, to address the scientific learning paradigm in the Industry 4.0 environment. Particular emphasis is placed on causal latent variable-based models using both data coming from a planned design of experiments and, mainly, data coming from the daily production process, namely happenstance data. The dissertation is structured in five parts. The first part discusses the scientific learning paradigm in the Industry 4.0 environment. The objectives of the thesis are highlighted. In addition to that, a comprehensive description of latent variable-based models is presented, on which the novel methodologies proposed in this thesis are founded. In the second part, the novel methodological contributions are presented. Firstly, the potential of PLS to analyze data from DOE, with or without missing runs is illustrated. Then, the potential of causal latent variable-based models is concentrated on defining the raw material design space providing assurance of quality with a certain confidence level for the critical to quality attributes, jointly with the development of a novel latent space-based multivariate capability index to rank and select suppliers for a particular raw material used in a manufacturing process. The third part aims to address novel applications by means of causal latent variable-based models using happenstance data. First, it concerns a health application: the Pandemic COVID-19. In this context, the use of latent variable-based models is applied to develop an alternative to placebo-controlled clinical trials. Then, latent variable-based models are used to optimize processes within the framework of industrial applications. The fourth part introduces a graphical user interface developed in Python code that integrates the developed methods with the aim of being self-explanatory and user-friendly. Finally, the last part discusses the relevance of this dissertation, including proposals that deserve further research. / Borràs Ferrís, J. (2023). Causal latent space-based models for scientific learning in Industry 4.0 [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/198993

Thesis submitted in fulfilment of the requirements for the degree Master of Technology Chemistry in the Faculty of (Science) Supervisor: Professor T.N. van der Walt Bellville campus Date submitted: October 2011 / It is important to quantify polymeric components in a coating because they greatly influence the performance of a coating. The difficulty associated with analysis of polymers by Fourier transform infrared (FTIR) analysis’s is that colinearities arise from similar or overlapping spectral features. A quantitative FTIR method with attenuated total reflectance coupled to multivariate/ chemometric analysis is presented. It allows for simultaneous quantification of 3 polymeric components; a rheology modifier, organic opacifier and styrene acrylic binder, with no prior extraction or separation from the paint. The factor based methods partial least squares (PLS) and principle component regression (PCR) permit colinearities by decomposing the spectral data into smaller matrices with principle scores and loading vectors. For model building spectral information from calibrators and validation samples at different analysis regions were incorporated. PCR and PLS were used to inspect the variation within the sample set. The PLS algorithms were found to predict the polymeric components the best. The concentrations of the polymeric components in a coating were predicted with the calibration model. Three PLS models each with different analysis regions yielded a coefficient of correlation R2 close to 1 for each of the components. The root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP) was less than 5%. The best out-put was obtained where spectral features of water was included (Trial 3). The prediction residual values for the three models ranged from 2 to -2 and 10 to -10. The method allows paint samples to be analysed in pure form and opens many opportunities for other coating components to be analysed in the same way.

En Analyse Sensorielle, les préférences des consommateurs sont généralement modélisées en fonction de données sensorielles par les méthodes de cartographie des préférences. L'objectif de cette thèse est de modéliser les préférences des consommateurs en intégrant, en plus des données sensorielles, de nouvelles variables relatives à leur perception des produits. Nous appellerons ces variables les dimensions subjectives. Elles recouvrent des dimensions pragmatiques liées à l'utilisation du produit et des dimensions plus symboliques telles que l'esthétisme, la modernité, l'originalité…Les problématiques relatives aux dimensions subjectives ont d'abord été étudiées lors d'une étude exploratoire sur des tasses à café. L'ensemble du travail a ensuite été réalisé sur 11 sièges de voitures. Dans un premier temps, des entretiens qualitatifs ont été réalisés auprès de 16 consommateurs d'une part et de 2 designers d'autre part. Ces entretiens ont permis d'identifier les dimensions subjectives caractéristiques des sièges. Une évaluation quantitative des dimensions subjectives et des préférences a ensuite été réalisée par 110 consommateurs. Enfin, les sièges ont été caractérisés sensoriellement par des experts. Les préférences des consommateurs ont été modélisées en fonction des données sensorielles et des dimensions subjectives par des modèles d'équations structurelles à variables latentes, plus précisément par Partial Least Square Path Modeling. Quatre modèles, fondés sur les groupes de préférences, ont été mis en place. Selon le groupe étudié, la contribution des deux jeux de données diffère et quatre profils de clients sont identifiés. D'un point de vue méthodologique, ce travail fournit des éléments de réponse sur l'intérêt des dimensions subjectives pour la modélisation des préférences. L'ensemble de la démarche est en cours d'application sur un produit alimentaire : le chocolat. / In Sensory Science, preference mapping is used to explain consumers' preferences with sensory data. This PhD aims to integrate not only sensory data but also new variables that are related to consumers' perception of the product in the modelling of consumers' preferences. These variables are labelled as subjective dimensions. They address the pragmatic dimensions that cover the context of use of the products and more symbolic dimensions, such as aesthetics, modernity, originality…An exploratory study based on coffee cups was a first mean to approach the issues related to subjective dimensions. Then, all the work was done on a study of 11 car seats. The first step consisted in qualitative interviews of 16 consumers and of 2 designers. These interviews allowed identifying the subjective dimensions that characterize car seats. 110 consumers then performed a quantitative evaluation of their preferences and subjective dimensions. Finally, the seats were characterized by experts with sensory descriptors. The consumers' preferences were modelled according to both sensory data and subjective dimensions, using structural equations: the Partial Least Square Path Modeling. Four models based on preferences clustering were established. The contribution of two kinds of data differed according to the considered cluster, which led to the identification of four customer profiles. From a methodological point of view, this work provides first elements about the benefit of subjective dimensions in preference modelling. The methodology is being implemented on a food product: chocolate.

This thesis challenges two key assumptions made in the current Islamic banking literature. Firstly, this thesis challenges and empirically invalidates the assumption that all Islamic banks are indistinguishable from their conventional counterparts and are thus equally unIslamic. To do so, this thesis uses the profit and loss sharing (PLS) criteria, which is central to the philosophy of Islamic banking and is the key principle differentiating Islamic from conventional banking, in theory and practice. By investigating variation in PLS levels between Islamic banks and comparing with conventional banks with and without Islamic windows, this thesis illustrates that the Islamic banking industry does not comprise a homogeneous group of banks that are all indistinguishable from their conventional counterparts. Rather, a typology of Islamic banks exists, comprising of three distinct groups of banks, each one following a different business model. While one group can genuinely be considered indistinguishable from conventional banks, another group shows clear evidence of pursuing PLS-oriented strategies in formulating its asset portfolio, differentiating itself from the purely debt-based intermediation model adopted by conventional banks. As such, empirical evidence shows that some Islamic banks are, in practice, operating closer to the PLS principle and can thus be considered more Islamic than others. Further investigation illustrates that the institutional environment matters for the provision of ideal PLS Islamic financing instruments. Secondly, this thesis overcomes two methodological issues to compare the corporate social performance (CSP) of Islamic and conventional banks. In doing so, this thesis challenges the second identified assumption from the literature, that religion-specific category of corporate social responsibility (CSR) is particular to Islamic banking, and invalidates it on conceptual, theoretical and empirical basis. A novel CSP Index based on the evidence-based disclosure criteria, comprising of 6 dimensions and 25 social performance indicators is constructed and complemented with three Social Performance Quantitative Indicators (SPQIs) to compare the CSP of Islamic and conventional banks. From this comparison, this thesis concludes that, contrary to the industry s claims and expectations held of it, Islamic banking does not offer an ethical alternative to conventional banking. Differences in the level and composition of CSP between the two industries are more subtle and require a nuanced approach to be studied.