Es werden differenzierbare zufällige Felder zweiter Ordnung untersucht und Vorschläge zur Versuchsplanung von Beobachtungen der abgeleiteten Felder unterbreitet. Von einem gewissen Standpunkt aus werden die folgenden Fragen beantwortet: Wie viele Informationen liefern Beobachtungen von Ableitungen für die Vorhersage des zugrunde liegenden Stochastischen Feldes? Wie beeinflusst eine a priori Wahl der Kovarianzfunktion das Informationsverhältnis zwischen verschiedenen abgeleiteten Feldern im Hinblick auf die Vorhersage? Als Zielfunktion wird das so genannte "imse-update" für den besten linearen Prädiktor betrachtet. Den zentralen Teil stellt die Untersuchung von Versuchsplänen mit (asymptotisch) verschwindenden Korrelationen dar. Hier wird insbesondere der Einfluss der Maternschen Klasse und J-Besselschen Klassen von Kovarianzfuntionen untersucht. Ferner wird der Einfluss gleichzeitiger Beobachtung von verschiedenen Ableitungen untersucht. Schließlich werden einige empirische Studien durchgeführt, aus denen einige praktische Ratschläge abgeleitet werden.

Im Rahmen dieser Arbeit wurde der Prozess des chemisch-mechanischen Polierens (CMP) von Siliziumwafern erstmals mittels stochastischer Methoden modelliert und daraus resultierend weiter optimiert. Ziel war es, Erkenntnisse zu ausgewählten, noch nicht vollständig verstandenen Einflussfaktoren zu gewinnen. Der Schwerpunkt lag dabei auf dem Poliertuch. Anhand eines neu entwickelten Modells zur Beschreibung einer konditionierten Tuchoberfläche wurden Zusammenhänge zwischen Konditionier- bzw. Tuchstrukturparametern und resultierender Poliertuchoberfläche herausgearbeitet und somit Möglichkeiten zur exakten Beschreibung und der gezielten Beeinflussung letzterer ermittelt. Weiterhin konnte erstmalig ein lang gesuchter messbarer Parameter benannt werden, mit dem eine ideale Tuchoberfläche charakterisierbar wird. Die Ergebnisse wurden experimentell verifiziert. Abschließend wurde mit einem neuen Abtragsmodell der CMP-Prozess von Siliziumwafern beschrieben, anhand dessen Zusammenhänge zwischen der Tuchrauheit und der Unebenheit der Waferoberfläche mit einer Theorie begründbar wurden.

Es werden differenzierbare zufällige Felder zweiter Ordnung untersucht und Vorschläge zur Versuchsplanung von Beobachtungen der abgeleiteten Felder unterbreitet. Von einem gewissen Standpunkt aus werden die folgenden Fragen beantwortet: Wie viele Informationen liefern Beobachtungen von Ableitungen für die Vorhersage des zugrunde liegenden Stochastischen Feldes? Wie beeinflusst eine a priori Wahl der Kovarianzfunktion das Informationsverhältnis zwischen verschiedenen abgeleiteten Feldern im Hinblick auf die Vorhersage? Als Zielfunktion wird das so genannte "imse-update" für den besten linearen Prädiktor betrachtet. Den zentralen Teil stellt die Untersuchung von Versuchsplänen mit (asymptotisch) verschwindenden Korrelationen dar. Hier wird insbesondere der Einfluss der Maternschen Klasse und J-Besselschen Klassen von Kovarianzfuntionen untersucht. Ferner wird der Einfluss gleichzeitiger Beobachtung von verschiedenen Ableitungen untersucht. Schließlich werden einige empirische Studien durchgeführt, aus denen einige praktische Ratschläge abgeleitet werden.

Im Rahmen dieser Arbeit wurde der Prozess des chemisch-mechanischen Polierens (CMP) von Siliziumwafern erstmals mittels stochastischer Methoden modelliert und daraus resultierend weiter optimiert. Ziel war es, Erkenntnisse zu ausgewählten, noch nicht vollständig verstandenen Einflussfaktoren zu gewinnen. Der Schwerpunkt lag dabei auf dem Poliertuch. Anhand eines neu entwickelten Modells zur Beschreibung einer konditionierten Tuchoberfläche wurden Zusammenhänge zwischen Konditionier- bzw. Tuchstrukturparametern und resultierender Poliertuchoberfläche herausgearbeitet und somit Möglichkeiten zur exakten Beschreibung und der gezielten Beeinflussung letzterer ermittelt. Weiterhin konnte erstmalig ein lang gesuchter messbarer Parameter benannt werden, mit dem eine ideale Tuchoberfläche charakterisierbar wird. Die Ergebnisse wurden experimentell verifiziert. Abschließend wurde mit einem neuen Abtragsmodell der CMP-Prozess von Siliziumwafern beschrieben, anhand dessen Zusammenhänge zwischen der Tuchrauheit und der Unebenheit der Waferoberfläche mit einer Theorie begründbar wurden.

The discretization of the stationary diffusion equation with random parameters by the Stochastic Finite Element Method requires the solution of a highly structured but very large linear system of equations. Depending on the stochastic properties of the diffusion coefficient together with the stochastic discretization we consider three solver cases. If the diffusion coefficient is given by a stochastically linear expansion, e.g. a truncated Karhunen-Loeve expansion, and tensor product polynomial stochastic shape functions are employed, the Galerkin matrix can be transformed to a block-diagonal matrix. For the solution of the resulting sequence of linear systems we study Krylov subspace recycling methods whose success depends on the ordering and grouping of the linear systems as well as the preconditioner. If we use complete polynomials for the stochastic discretization instead, we show that decoupling of the Galerkin matrix with respect to the stochastic degrees of freedom is impossible. For a stochastically nonlinear diffusion coefficient, e.g. a lognormal random field, together with complete polynomials serving as stochastic shape functions, we introduce and test the performance of a new Kronecker product preconditioner, which is not exclusively based on the mean value of the diffusion coefficient.

The discretization of the stationary diffusion equation with random parameters by the Stochastic Finite Element Method requires the solution of a highly structured but very large linear system of equations. Depending on the stochastic properties of the diffusion coefficient together with the stochastic discretization we consider three solver cases. If the diffusion coefficient is given by a stochastically linear expansion, e.g. a truncated Karhunen-Loeve expansion, and tensor product polynomial stochastic shape functions are employed, the Galerkin matrix can be transformed to a block-diagonal matrix. For the solution of the resulting sequence of linear systems we study Krylov subspace recycling methods whose success depends on the ordering and grouping of the linear systems as well as the preconditioner. If we use complete polynomials for the stochastic discretization instead, we show that decoupling of the Galerkin matrix with respect to the stochastic degrees of freedom is impossible. For a stochastically nonlinear diffusion coefficient, e.g. a lognormal random field, together with complete polynomials serving as stochastic shape functions, we introduce and test the performance of a new Kronecker product preconditioner, which is not exclusively based on the mean value of the diffusion coefficient.

In geotechnical engineering, the uncertainties such as the variability and uncertainty inherent in the geotechnical properties have caught more and more attentions from researchers and engineers. They have found that a single “Factor of Safety” calculated by traditional deterministic analyses methods can not represent the slope stability exactly. Recently in order to provide a more rational mathematical framework to incorporate different types of uncertainties in the slope stability estimation, reliability analyses and non-deterministic methods, which include probabilistic and non probabilistic (imprecise methods) methods, have been applied widely. In short, the slope non-deterministic analysis is to combine the probabilistic analysis or non probabilistic analysis with the deterministic slope stability analysis. It cannot be regarded as a completely new slope stability analysis method, but just an extension of the slope deterministic analysis. The slope failure probability calculated by slope non-deterministic analysis is a kind of complement of safety factor. Therefore, the accuracy of non deterministic analysis is not only depended on a suitable probabilistic or non probabilistic analysis method selected, but also on a more rigorous deterministic analysis method or geological model adopted. In this thesis, reliability concepts have been reviewed first, and some typical non-deterministic methods, including Monte Carlo Simulation (MCS), First Order Reliability Method (FORM), Point Estimate Method (PEM) and Random Set Theory (RSM), have been described and successfully applied to the slope stability analysis based on a numerical simulation method-Strength Reduction Method (SRM). All of the processes have been performed in a commercial finite difference code FLAC and a distinct element code UDEC. First of all, as the fundamental of slope reliability analysis, the deterministic numerical simulation method has been improved. This method has a higher accuracy than the conventional limit equilibrium methods, because of the reason that the constitutive relationship of soil is considered, and fewer assumptions on boundary conditions of slope model are necessary. However, the construction of slope numerical models, particularly for the large and complicated models has always been very difficult and it has become an obstacle for application of numerical simulation method. In this study, the excellent spatial analysis function of Geographic Information System (GIS) technique has been introduced to help numerical modeling of the slope. In the process of modeling, the topographic map of slope has been gridded using GIS software, and then the GIS data was transformed into FLAC smoothly through the program built-in language FISH. At last, the feasibility and high efficiency of this technique has been illustrated through a case study-Xuecheng slope, and both 2D and 3D models have been investigated. Subsequently, three most widely used probabilistic analyses methods, Monte Carlo Simulation, First Order Reliability Method and Point Estimate Method applied with Strength Reduction Method have been studied. Monte Carlo Simulation which needs to repeat thousands of deterministic analysis is the most accurate probabilistic method. However it is too time consuming for practical applications, especially when it is combined with numerical simulation method. For reducing the computation effort, a simplified Monte Carlo Simulation-Strength Reduction Method (MCS-SRM) has been developed in this study. This method has estimated the probable failure of slope and calculated the mean value of safety factor by means of soil parameters first, and then calculated the variance of safety factor and reliability of slope according to the assumed probability density function of safety factor. Case studies have confirmed that this method can reduce about 4/5 of time compared with traditional MCS-SRM, and maintain almost the same accuracy. First Order Reliability Method is an approximate method which is based on the Taylor\'s series expansion of performance function. The closed form solution of the partial derivatives of the performance function is necessary to calculate the mean and standard deviation of safety factor. However, there is no explicit performance function in numerical simulation method, so the derivative expressions have been replaced with equivalent difference quotients to solve the differential quotients approximately in this study. Point Estimate Method is also an approximate method involved even fewer calculations than FORM. In the present study, it has been integrated with Strength Reduction Method directly. Another important observation referred to the correlation between the soil parameters cohesion and friction angle. Some authors have found a negative correlation between cohesion and friction angle of soil on the basis of experimental data. However, few slope probabilistic studies are found to consider this negative correlation between soil parameters in literatures. In this thesis, the influence of this correlation on slope probability of failure has been investigated based on numerical simulation method. It was found that a negative correlation considered in the cohesion and friction angle of soil can reduce the variability of safety factor and failure probability of slope, thus increasing the reliability of results. Besides inter-correlation of soil parameters, these are always auto-correlated in space, which is described as spatial variability. For the reason that knowledge on this character is rather limited in literature, it is ignored in geotechnical engineering by most researchers and engineers. In this thesis, the random field method has been introduced in slope numerical simulation to simulate the spatial variability structure, and a numerical procedure for a probabilistic slope stability analysis based on Monte Carlo simulation was presented. The soil properties such as cohesion and friction angle were discretized to continuous random fields based on local averaging method. In the case study, both stationary and non-stationary random fields have been investigated, and the influence of spatial variability and averaging domain on the convergence of numerical simulation and probability of failure was studied. In rock medium, the structure faces have very important influence on the slope stability, and the rock material can be modeled as the combination of rigid or deformable blocks with joints in distinct element method. Therefore, much more input parameters like strength of joints are required to input the rock slope model, which increase the uncertainty of the results of numerical model. Furthermore, because of the limitations of the current laboratory and in-site testes, there is always lack of exact values of geotechnical parameters from rock material, even the probability distribution of these variables. Most of time, engineers can only estimate the interval of these variables from the limit testes or the expertise’s experience. In this study, to assess the reliability of the rock slope, a Random Set Distinct Element Method (RS-DEM) has been developed through coupling of Random Set Theory and Distinct Element Method, and applied in a rock slope in Sichuan province China.

In geotechnical engineering, the uncertainties such as the variability and uncertainty inherent in the geotechnical properties have caught more and more attentions from researchers and engineers. They have found that a single “Factor of Safety” calculated by traditional deterministic analyses methods can not represent the slope stability exactly. Recently in order to provide a more rational mathematical framework to incorporate different types of uncertainties in the slope stability estimation, reliability analyses and non-deterministic methods, which include probabilistic and non probabilistic (imprecise methods) methods, have been applied widely. In short, the slope non-deterministic analysis is to combine the probabilistic analysis or non probabilistic analysis with the deterministic slope stability analysis. It cannot be regarded as a completely new slope stability analysis method, but just an extension of the slope deterministic analysis. The slope failure probability calculated by slope non-deterministic analysis is a kind of complement of safety factor. Therefore, the accuracy of non deterministic analysis is not only depended on a suitable probabilistic or non probabilistic analysis method selected, but also on a more rigorous deterministic analysis method or geological model adopted. In this thesis, reliability concepts have been reviewed first, and some typical non-deterministic methods, including Monte Carlo Simulation (MCS), First Order Reliability Method (FORM), Point Estimate Method (PEM) and Random Set Theory (RSM), have been described and successfully applied to the slope stability analysis based on a numerical simulation method-Strength Reduction Method (SRM). All of the processes have been performed in a commercial finite difference code FLAC and a distinct element code UDEC. First of all, as the fundamental of slope reliability analysis, the deterministic numerical simulation method has been improved. This method has a higher accuracy than the conventional limit equilibrium methods, because of the reason that the constitutive relationship of soil is considered, and fewer assumptions on boundary conditions of slope model are necessary. However, the construction of slope numerical models, particularly for the large and complicated models has always been very difficult and it has become an obstacle for application of numerical simulation method. In this study, the excellent spatial analysis function of Geographic Information System (GIS) technique has been introduced to help numerical modeling of the slope. In the process of modeling, the topographic map of slope has been gridded using GIS software, and then the GIS data was transformed into FLAC smoothly through the program built-in language FISH. At last, the feasibility and high efficiency of this technique has been illustrated through a case study-Xuecheng slope, and both 2D and 3D models have been investigated. Subsequently, three most widely used probabilistic analyses methods, Monte Carlo Simulation, First Order Reliability Method and Point Estimate Method applied with Strength Reduction Method have been studied. Monte Carlo Simulation which needs to repeat thousands of deterministic analysis is the most accurate probabilistic method. However it is too time consuming for practical applications, especially when it is combined with numerical simulation method. For reducing the computation effort, a simplified Monte Carlo Simulation-Strength Reduction Method (MCS-SRM) has been developed in this study. This method has estimated the probable failure of slope and calculated the mean value of safety factor by means of soil parameters first, and then calculated the variance of safety factor and reliability of slope according to the assumed probability density function of safety factor. Case studies have confirmed that this method can reduce about 4/5 of time compared with traditional MCS-SRM, and maintain almost the same accuracy. First Order Reliability Method is an approximate method which is based on the Taylor\'s series expansion of performance function. The closed form solution of the partial derivatives of the performance function is necessary to calculate the mean and standard deviation of safety factor. However, there is no explicit performance function in numerical simulation method, so the derivative expressions have been replaced with equivalent difference quotients to solve the differential quotients approximately in this study. Point Estimate Method is also an approximate method involved even fewer calculations than FORM. In the present study, it has been integrated with Strength Reduction Method directly. Another important observation referred to the correlation between the soil parameters cohesion and friction angle. Some authors have found a negative correlation between cohesion and friction angle of soil on the basis of experimental data. However, few slope probabilistic studies are found to consider this negative correlation between soil parameters in literatures. In this thesis, the influence of this correlation on slope probability of failure has been investigated based on numerical simulation method. It was found that a negative correlation considered in the cohesion and friction angle of soil can reduce the variability of safety factor and failure probability of slope, thus increasing the reliability of results. Besides inter-correlation of soil parameters, these are always auto-correlated in space, which is described as spatial variability. For the reason that knowledge on this character is rather limited in literature, it is ignored in geotechnical engineering by most researchers and engineers. In this thesis, the random field method has been introduced in slope numerical simulation to simulate the spatial variability structure, and a numerical procedure for a probabilistic slope stability analysis based on Monte Carlo simulation was presented. The soil properties such as cohesion and friction angle were discretized to continuous random fields based on local averaging method. In the case study, both stationary and non-stationary random fields have been investigated, and the influence of spatial variability and averaging domain on the convergence of numerical simulation and probability of failure was studied. In rock medium, the structure faces have very important influence on the slope stability, and the rock material can be modeled as the combination of rigid or deformable blocks with joints in distinct element method. Therefore, much more input parameters like strength of joints are required to input the rock slope model, which increase the uncertainty of the results of numerical model. Furthermore, because of the limitations of the current laboratory and in-site testes, there is always lack of exact values of geotechnical parameters from rock material, even the probability distribution of these variables. Most of time, engineers can only estimate the interval of these variables from the limit testes or the expertise’s experience. In this study, to assess the reliability of the rock slope, a Random Set Distinct Element Method (RS-DEM) has been developed through coupling of Random Set Theory and Distinct Element Method, and applied in a rock slope in Sichuan province China.