Desarrollo de un Modelo de Gestión de Riesgos Operacionales basado en las Directrices del Comité de Basilea

La presente investigación tiene como objetivo principal diseñar un Modelo de Gestión de Riesgos Operacionales (MGRO) según las Directrices de los Acuerdos II y III del Comité de Supervisión Bancaria de Basilea del Banco de Pagos Internacionales (CSBB-BPI). Se considera importante realizar un estudio sobre este tema dado que son los riesgos operacionales (OpR) los responsables en gran medida de las últimas crisis financieras mundiales y por la dificultad para detectarlos en las organizaciones. Se ha planteado un modelo de gestión subdividido en dos vías de influencias. La primera acoge el paradigma holístico en el que se considera que hay múltiples maneras de percibir un proceso cíclico, así como las herramientas para observar, conocer y entender el objeto o sujeto percibido. La segunda vía la representa el paradigma totalizante, en el que se obtienen datos tanto cualitativos como cuantitativos, los cuales son complementarios entre si. Por otra parte, este trabajo plantea el diseño de un programa informático de OpR Cualitativo, que ha sido diseñado para determinar la raíz de los riesgos en las organizaciones y su Valor en Riesgo Operacional (OpVaR) basado en el método del indicador básico. Aplicando el ciclo holístico al caso de estudio, se obtuvo el siguiente diseño de investigación: no experimental, univariable, transversal descriptiva, contemporánea, retrospectiva, de fuente mixta, cualitativa (fenomenológica y etnográfica) y cuantitativa (descriptiva y analítica). La toma de decisiones y recolección de información se realizó en dos fases en la unidad de estudio. En la primera se tomó en cuenta la totalidad de la empresa Corpoelec-EDELCA, en la que se presentó un universo estadístico de 4271 personas, una población de 2390 personas y una unidad de muestreo de 87 personas. Se repitió el proceso en una segunda fase, para la Central Hidroeléctrica Simón Bolívar, y se determinó un segundo universo estadístico de 300 trabajadores, una población de 191 personas y una muestra de 58 profesionales. Como fuentes de recolección de información se utilizaron fuentes primarias y secundarias. Para recabar la información primaria se realizaron observaciones directas, dos encuestas para detectar las áreas y procesos con mayor nivel de riesgos y se diseñó un cuestionario combinado con otra encuesta (ad hoc) para establecer las estimaciones de frecuencia y severidad de pérdidas operacionales. La información de fuentes secundarias se extrajo de las bases de datos de Corpoelec-EDELCA, de la IEA, del Banco Mundial, del CSBB-BPI, de la UPM y de la UC at Berkeley, entre otras. Se establecieron las distribuciones de frecuencia y de severidad de pérdidas operacionales como las variables independientes y el OpVaR como la variable dependiente. No se realizó ningún tipo de seguimiento o control a las variables bajo análisis, ya que se consideraron estas para un instante especifico y solo se determinan con la finalidad de establecer la existencia y valoración puntual de los OpR en la unidad de estudio. El análisis cualitativo planteado en el MGRO, permitió detectar que en la unidad de investigación, el 67% de los OpR detectados provienen de dos fuentes principales: procesos (32%) y eventos externos (35%). Adicionalmente, la validación del MGRO en Corpoelec-EDELCA, permitió detectar que el 63% de los OpR en la organización provienen de tres categorías principales, siendo los fraudes externos los presentes con mayor regularidad y severidad de pérdidas en la organización. La exposición al riesgo se determinó fundamentándose en la adaptación del concepto de OpVaR que generalmente se utiliza para series temporales y que en el caso de estudio presenta la primicia de aplicarlo a datos cualitativos transformados con la escala Likert. La posibilidad de utilizar distribuciones de probabilidad típicas para datos cuantitativos en distribuciones de frecuencia y severidad de pérdidas con datos de origen cualitativo fueron analizadas. Para el 64% de los OpR estudiados se obtuvo que la frecuencia tiene un comportamiento semejante al de la distribución de probabilidad de Poisson y en un 55% de los casos para la severidad de pérdidas se obtuvo a las log-normal como las distribuciones de probabilidad más comunes, con lo que se concluyó que los enfoques sugeridos por el BCBS-BIS para series de tiempo son aplicables a los datos cualitativos. Obtenidas las distribuciones de frecuencia y severidad de pérdidas, se convolucionaron estas implementando el método de Montecarlo, con lo que se obtuvieron los enfoques de distribuciones de pérdidas (LDA) para cada uno de los OpR. El OpVaR se dedujo como lo sugiere el CSBB-BPI del percentil 99,9 o 99% de cada una de las LDA, obteniéndose que los OpR presentan un comportamiento similar al sistema financiero, resultando como los de mayor peligrosidad los que se ubican con baja frecuencia y alto impacto, por su dificultad para ser detectados y monitoreados. Finalmente, se considera que el MGRO permitirá a los agentes del mercado y sus grupos de interés conocer con efectividad, fiabilidad y eficiencia el status de sus entidades, lo que reducirá la incertidumbre de sus inversiones y les permitirá establecer una nueva cultura de gestión en sus organizaciones. ABSTRACT This research has as main objective the design of a Model for Operational Risk Management (MORM) according to the guidelines of Accords II and III of the Basel Committee on Banking Supervision of the Bank for International Settlements (BCBS- BIS). It is considered important to conduct a study on this issue since operational risks (OpR) are largely responsible for the recent world financial crisis and due to the difficulty in detecting them in organizations. A management model has been designed which is divided into two way of influences. The first supports the holistic paradigm in which it is considered that there are multiple ways of perceiving a cyclical process and contains the tools to observe, know and understand the subject or object perceived. The second way is the totalizing paradigm, in which both qualitative and quantitative data are obtained, which are complementary to each other. Moreover, this paper presents the design of qualitative OpR software which is designed to determine the root of risks in organizations and their Operational Value at Risk (OpVaR) based on the basic indicator approach. Applying the holistic cycle to the case study, the following research design was obtained: non- experimental, univariate, descriptive cross-sectional, contemporary, retrospective, mixed-source, qualitative (phenomenological and ethnographic) and quantitative (descriptive and analytical). Decision making and data collection was conducted in two phases in the study unit. The first took into account the totality of the Corpoelec-EDELCA company, which presented a statistical universe of 4271 individuals, a population of 2390 individuals and a sampling unit of 87 individuals. The process was repeated in a second phase to the Simon Bolivar Hydroelectric Power Plant, and a second statistical universe of 300 workers, a population of 191 people and a sample of 58 professionals was determined. As sources of information gathering primary and secondary sources were used. To obtain the primary information direct observations were conducted and two surveys to identify the areas and processes with higher risks were designed. A questionnaire was combined with an ad hoc survey to establish estimates of frequency and severity of operational losses was also considered. The secondary information was extracted from the databases of Corpoelec-EDELCA, IEA, the World Bank, the BCBS-BIS, UPM and UC at Berkeley, among others. The operational loss frequency distributions and the operational loss severity distributions were established as the independent variables and OpVaR as the dependent variable. No monitoring or control of the variables under analysis was performed, as these were considered for a specific time and are determined only for the purpose of establishing the existence and timely assessment of the OpR in the study unit. Qualitative analysis raised in the MORM made it possible to detect that in the research unit, 67% of detected OpR come from two main sources: external processes (32%) and external events (35%). Additionally, validation of the MORM in Corpoelec-EDELCA, enabled to estimate that 63% of OpR in the organization come from three main categories, with external fraud being present more regularly and greater severity of losses in the organization. Risk exposure is determined basing on adapting the concept of OpVaR generally used for time series and in the case study it presents the advantage of applying it to qualitative data transformed with the Likert scale. The possibility of using typical probability distributions for quantitative data in loss frequency and loss severity distributions with data of qualitative origin were analyzed. For the 64% of OpR studied it was found that the frequency has a similar behavior to that of the Poisson probability distribution and 55% of the cases for loss severity it was found that the log-normal were the most common probability distributions. It was concluded that the approach suggested by the BCBS-BIS for time series can be applied to qualitative data. Once obtained the distributions of loss frequency and severity have been obtained they were subjected to convolution implementing the Monte Carlo method. Thus the loss distribution approaches (LDA) were obtained for each of the OpR. The OpVaR was derived as suggested by the BCBS-BIS 99.9 percentile or 99% of each of the LDA. It was determined that the OpR exhibits a similar behavior to the financial system, being the most dangerous those with low frequency and high impact for their difficulty in being detected and monitored. Finally, it is considered that the MORM will allows market players and their stakeholders to know with effectiveness, efficiency and reliability the status of their entities, which will reduce the uncertainty of their investments and enable them to establish a new management culture in their organizations.

Modelo de gestión de un embalse en tiempo real durante avenidas basado en redes bayesianas entrenadas con el método de optimización PLEM

En la actualidad, la gestión de embalses para el control de avenidas se realiza, comúnmente, utilizando modelos de simulación. Esto se debe, principalmente, a su facilidad de uso en tiempo real por parte del operador de la presa. Se han desarrollado modelos de optimización de la gestión del embalse que, aunque mejoran los resultados de los modelos de simulación, su aplicación en tiempo real se hace muy difícil o simplemente inviable, pues está limitada al conocimiento de la avenida futura que entra al embalse antes de tomar la decisión de vertido. Por esta razón, se ha planteado el objetivo de desarrollar un modelo de gestión de embalses en avenidas que incorpore las ventajas de un modelo de optimización y que sea de fácil uso en tiempo real por parte del gestor de la presa. Para ello, se construyó un modelo de red Bayesiana que representa los procesos de la cuenca vertiente y del embalse y, que aprende de casos generados sintéticamente mediante un modelo hidrológico agregado y un modelo de optimización de la gestión del embalse. En una primera etapa, se generó un gran número de episodios sintéticos de avenida utilizando el método de Monte Carlo, para obtener las lluvias, y un modelo agregado compuesto de transformación lluvia- escorrentía, para obtener los hidrogramas de avenida. Posteriormente, se utilizaron las series obtenidas como señales de entrada al modelo de gestión de embalses PLEM, que optimiza una función objetivo de costes mediante programación lineal entera mixta, generando igual número de eventos óptimos de caudal vertido y de evolución de niveles en el embalse. Los episodios simulados fueron usados para entrenar y evaluar dos modelos de red Bayesiana, uno que pronostica el caudal de entrada al embalse, y otro que predice el caudal vertido, ambos en un horizonte de tiempo que va desde una a cinco horas, en intervalos de una hora. En el caso de la red Bayesiana hidrológica, el caudal de entrada que se elige es el promedio de la distribución de probabilidad de pronóstico. En el caso de la red Bayesiana hidráulica, debido al comportamiento marcadamente no lineal de este proceso y a que la red Bayesiana devuelve un rango de posibles valores de caudal vertido, se ha desarrollado una metodología para seleccionar un único valor, que facilite el trabajo del operador de la presa. Esta metodología consiste en probar diversas estrategias propuestas, que incluyen zonificaciones y alternativas de selección de un único valor de caudal vertido en cada zonificación, a un conjunto suficiente de episodios sintéticos. Los resultados de cada estrategia se compararon con el método MEV, seleccionándose las estrategias que mejoran los resultados del MEV, en cuanto al caudal máximo vertido y el nivel máximo alcanzado por el embalse, cualquiera de las cuales puede usarse por el operador de la presa en tiempo real para el embalse de estudio (Talave). La metodología propuesta podría aplicarse a cualquier embalse aislado y, de esta manera, obtener, para ese embalse particular, diversas estrategias que mejoran los resultados del MEV. Finalmente, a modo de ejemplo, se ha aplicado la metodología a una avenida sintética, obteniendo el caudal vertido y el nivel del embalse en cada intervalo de tiempo, y se ha aplicado el modelo MIGEL para obtener en cada instante la configuración de apertura de los órganos de desagüe que evacuarán el caudal. Currently, the dam operator for the management of dams uses simulation models during flood events, mainly due to its ease of use in real time. Some models have been developed to optimize the management of the reservoir to improve the results of simulation models. However, real-time application becomes very difficult or simply unworkable, because the decision to discharge depends on the unknown future avenue entering the reservoir. For this reason, the main goal is to develop a model of reservoir management at avenues that incorporates the advantages of an optimization model. At the same time, it should be easy to use in real-time by the dam manager. For this purpose, a Bayesian network model has been developed to represent the processes of the watershed and reservoir. This model learns from cases generated synthetically by a hydrological model and an optimization model for managing the reservoir. In a first stage, a large number of synthetic flood events was generated using the Monte Carlo method, for rain, and rain-added processing model composed of runoff for the flood hydrographs. Subsequently, the series obtained were used as input signals to the reservoir management model PLEM that optimizes a target cost function using mixed integer linear programming. As a result, many optimal discharge rate events and water levels in the reservoir levels were generated. The simulated events were used to train and test two models of Bayesian network. The first one predicts the flow into the reservoir, and the second predicts the discharge flow. They work in a time horizon ranging from one to five hours, in intervals of an hour. In the case of hydrological Bayesian network, the chosen inflow is the average of the probability distribution forecast. In the case of hydraulic Bayesian network the highly non-linear behavior of this process results on a range of possible values of discharge flow. A methodology to select a single value has been developed to facilitate the dam operator work. This methodology tests various strategies proposed. They include zoning and alternative selection of a single value in each discharge rate zoning from a sufficient set of synthetic episodes. The results of each strategy are compared with the MEV method. The strategies that improve the outcomes of MEV are selected and can be used by the dam operator in real time applied to the reservoir study case (Talave). The methodology could be applied to any single reservoir and, thus, obtain, for the particular reservoir, various strategies that improve results from MEV. Finally, the methodology has been applied to a synthetic flood, obtaining the discharge flow and the reservoir level in each time interval. The open configuration floodgates to evacuate the flow at each interval have been obtained applying the MIGEL model.

Análisis paramétrico determinista y aleatorio de problemas dinámicos en estructuras aeroespaciales

Los fenómenos dinámicos pueden poner en peligro la integridad de estructuras aeroespaciales y los ingenieros han desarrollado diferentes estrategias para analizarlos. Uno de los grandes problemas que se plantean en la ingeniería es cómo atacar un problema dinámico estructural. En la presente tesis se plantean distintos fenómenos dinámicos y se proponen métodos para estimar o simular sus comportamientos mediante un análisis paramétrico determinista y aleatorio del problema. Se han propuesto desde problemas sencillos con pocos grados de libertad que sirven para analizar las diferentes estrategias y herramientas a utilizar, hasta fenómenos muy dinámicos que contienen comportamientos no lineales, daños y fallos. Los primeros ejemplos de investigación planteados cubren una amplia gama de los fenómenos dinámicos, como el análisis de vibraciones de elementos másicos, incluyendo impactos y contactos, y el análisis de una viga con carga armónica aplicada a la que también se le añaden parámetros aleatorios que pueden responder a un desconocimiento o incertidumbre de los mismos. Durante el desarrollo de la tesis se introducen conceptos y se aplican distintos métodos, como el método de elementos finitos (FEM) en el que se analiza su resolución tanto por esquemas implícitos como explícitos, y métodos de análisis paramétricos y estadísticos mediante la técnica de Monte Carlo. Más adelante, una vez ya planteadas las herramientas y estrategias de análisis, se estudian fenómenos más complejos, como el impacto a baja velocidad en materiales compuestos, en el que se busca evaluar la resistencia residual y, por lo tanto, la tolerancia al daño de la estructura. Se trata de un suceso que puede producirse por la caída de herramienta, granizo o restos en la pista de aterrizaje. Otro de los fenómenos analizados también se da en un aeropuerto y se trata de la colisión con un dispositivo frangible, el cual tiene que romperse bajo ciertas cargas y, sin embargo, soportar otras. Finalmente, se aplica toda la metodología planteada en simular y analizar un posible incidente en vuelo, el fenómeno de la pérdida de pala de un turbohélice. Se trata de un suceso muy particular en el que la estructura tiene que soportar unas cargas complejas y excepcionales con las que la aeronave debe ser capaz de completar con éxito el vuelo. El análisis incluye comportamientos no lineales, daños, y varios tipos de fallos, y en el que se trata de identificar los parámetros clave en la secuencia del fallo. El suceso se analiza mediante análisis estructurales deterministas más habituales y también mediante otras técnicas como el método de Monte Carlo con el que se logran estudiar distintas incertidumbres en los parámetros con variables aleatorias. Se estudian, entre otros, el tamaño de pala perdida, la velocidad y el momento en el que se produce la rotura, y la rigidez y resistencia de los apoyos del motor. Se tiene en cuenta incluso el amortiguamiento estructural del sistema. Las distintas estrategias de análisis permiten obtener unos resultados valiosos e interesantes que han sido objeto de distintas publicaciones. ABSTRACT Dynamic phenomena can endanger the integrity of aerospace structures and, consequently, engineers have developed different strategies to analyze them. One of the major engineering problems is how to deal with the structural dynamics. In this thesis, different dynamic phenomena are introduced and several methods are proposed to estimate or simulate their behaviors. The analysis is considered through parametric, deterministic and statistical methods. The suggested issues are from simple problems with few degrees of freedom, in order to develop different strategies and tools to solve them, to very dynamic phenomena containing nonlinear behaviors failures, damages. The first examples cover a wide variety of dynamic phenomena such as vibration analysis of mass elements, including impacts and contacts, and beam analysis with harmonic load applied, in which random parameters are included. These parameters can represent the unawareness or uncertainty of certain variables. During the development of the thesis several concepts are introduced and different methods are applied, such as the finite element method (FEM), which is solved through implicit and explicit schemes, and parametrical and statistical methods using the Monte Carlo analysis technique. Next, once the tools and strategies of analysis are set out more complex phenomena are studied. This is the case of a low-speed impact in composite materials, the residual strength of the structure is evaluated, and therefore, its damage tolerance. This incident may occur from a tool dropped, hail or debris throw on the runway. At an airport may also occur, and it is also analyzed, a collision between an airplane and a frangible device. The devise must brake under these loads, however, it must withstand others. Finally, all the considered methodology is applied to simulate and analyze a flight incident, the blade loss phenomenon of a turboprop. In this particular event the structure must support complex and exceptional loads and the aircraft must be able to successfully complete the flight. Nonlinear behavior, damage, and different types of failures are included in the analysis, in which the key parameters in the failure sequence are identified. The incident is analyzed by deterministic structural analysis and also by other techniques such as Monte Carlo method, in which it is possible to include different parametric uncertainties through random variables. Some of the evaluated parameters are, among others, the blade loss size, propeller rotational frequency, speed and angular position where the blade is lost, and the stiffness and strength of the engine mounts. The study does also research on the structural damping of the system. The different strategies of analysis obtain valuable and interesting results that have been already published.

Magnetic activity and differential rotation in the young Sun-like stars KIC 7985370 and KIC 7765135

Aims. We present a detailed study of the two Sun-like stars KIC 7985370 and KIC 7765135, to determine their activity level, spot distribution, and differential rotation. Both stars were previously discovered by us to be young stars and were observed by the NASA Kepler mission. Methods. The fundamental stellar parameters (vsini, spectral type, T_eff, log g, and [Fe/H]) were derived from optical spectroscopy by comparison with both standard-star and synthetic spectra. The spectra of the targets allowed us to study the chromospheric activity based on the emission in the core of hydrogen Hα and Ca ii infrared triplet (IRT) lines, which was revealed by the subtraction of inactive templates. The high-precision Kepler photometric data spanning over 229 days were then fitted with a robust spot model. Model selection and parameter estimation were performed in a Bayesian manner, using a Markov chain Monte Carlo method. Results. We find that both stars are Sun-like (of G1.5 V spectral type) and have an age of about 100–200 Myr, based on their lithium content and kinematics. Their youth is confirmed by their high level of chromospheric activity, which is comparable to that displayed by the early G-type stars in the Pleiades cluster. The Balmer decrement and flux ratio of their Ca ii-IRT lines suggest that the formation of the core of these lines occurs mainly in optically thick regions that are analogous to solar plages. The spot model applied to the Kepler photometry requires at least seven persistent spots in the case of KIC 7985370 and nine spots in the case of KIC 7765135 to provide a satisfactory fit to the data. The assumption of the longevity of the star spots, whose area is allowed to evolve with time, is at the heart of our spot-modelling approach. On both stars, the surface differential rotation is Sun-like, with the high-latitude spots rotating slower than the low-latitude ones. We found, for both stars, a rather high value of the equator-to-pole differential rotation (dΩ ≈ 0.18 rad d^-1), which disagrees with the predictions of some mean-field models of differential rotation for rapidly rotating stars. Our results agree instead with previous works on solar-type stars and other models that predict a higher latitudinal shear, increasing with equatorial angular velocity, that can vary during the magnetic cycle.

Confiabilidade aplicada ao problema de interação estaca-solo.

Este trabalho busca aplicar técnicas de confiabilidade ao problema de grupo de estacas utilizadas como fundação de estruturas correntes. Para isso, lança-se mão de um modelo tridimensional de interação estaca-solo onde estão presentes o Método dos Elementos de Contorno (MEC) e o método dos Elementos Finitos (MEF) que atuam de forma acoplada. O MEC, com as soluções fundamentais de Mindlin (meio semi-infinito, homogêneo, isotrópico e elástico-linear é utiliza), é utilizado para modelar o solo. Já o MEF é utilizado para modelar as estacas. Definido o modelo de funcionamento estrutural das estacas, parte-se para a aplicação de métodos trazidos da confiabilidade estrutural para avaliação da adequabilidade em relação aos estados limite de serviço e estados limites últimos. Os métodos de confiabilidade utilizados foram o Método de Monte Carlo, o método FOSM (First-Order Second-Moment) e o método FORM (First-Order Reliability Method).

First-order transition in a three-dimensional disordered system

We present the first detailed numerical study in three dimensions of a first-order phase transition that remains first order in the presence of quenched disorder (specifically, the ferromagnetic-paramagnetic transition of the site-diluted four states Potts model). A tricritical point, which lies surprisingly near the pure-system limit and is studied by means of finite-size scaling, separates the first-order and second-order parts of the critical line. This investigation has been made possible by a new definition of the disorder average that avoids the diverging-variance probability distributions that plague the standard approach. Entropy, rather than free energy, is the basic object in this approach that exploits a recently introduced microcanonical Monte Carlo method.

First Order Phase Transition in a 3D disordered system

We present a detailed numerical study on the effects of adding quenched impurities to a three dimensional system which in the pure case undergoes a strong first order phase transition (specifically, the ferromagnetic/paramagnetic transition of the site-diluted four states Potts model). We can state that the transition remains first-order in the presence of quenched disorder (a small amount of it) but it turns out to be second order as more impurities are added. A tricritical point, which is studied by means of Finite-Size Scaling, separates the first-order and second-order parts of the critical line. The results were made possible by a new definition of the disorder average that avoids the diverging-variance probability distributions that arise using the standard methodology. We also made use of a recently proposed microcanonical Monte Carlo method in which entropy, instead of free energy, is the basic quantity.

Efeito da corrosão do chapeamento do fundo do casco sobre a confiabilidade estrutural de navios.

A engenharia é a ciência que transforma os conhecimentos das disciplinas básicas aplicadas a fatos reais. Nosso mundo está rodeado por essas realizações da engenharia, e é necessário que as pessoas se sintam confortáveis e seguras nas mesmas. Assim, a segurança se torna um fator importante que deve ser considerado em qualquer projeto. Na engenharia naval, um apropriado nível de segurança e, em consequência, um correto desenho estrutural é baseado, atualmente, em estudos determinísticos com o objetivo de obter estruturas capazes de suportar o pior cenário possível de solicitações durante um período de tempo determinado. A maior parte das solicitações na estrutura de um navio se deve à ação da natureza (ventos, ondas, correnteza e tempestades), ou, ainda, aos erros cometidos por humanos (explosões internas, explosões externas e colisões). Devido à aleatoriedade destes eventos, a confiabilidade estrutural de um navio deveria ser considerada como um problema estocástico sob condições ambientais bem caracterizadas. A metodologia probabilística, baseada em estatística e incertezas, oferece uma melhor perspectiva dos fenômenos reais que acontecem na estrutura dos navios. Esta pesquisa tem como objetivo apresentar resultados de confiabilidade estrutural em projetos e planejamento da manutenção para a chapa do fundo dos cascos dos navios, as quais são submetidas a esforços variáveis pela ação das ondas do mar e da corrosão. Foram estudados modelos estatísticos para a avaliação da estrutura da viga-navio e para o detalhe estrutural da chapa do fundo. Na avaliação da estrutura da viga-navio, o modelo desenvolvido consiste em determinar as probabilidades de ocorrência das solicitações na estrutura, considerando a deterioração por corrosão, com base numa investigação estatística da variação dos esforços em função das ondas e a deterioração em função de uma taxa de corrosão padrão recomendada pela DET NORSKE VERITAS (DNV). A abordagem para avaliação da confiabilidade dependente do tempo é desenvolvida com base nas curvas de resistências e solicitações (R-S) determinadas pela utilização do método de Monte Carlo. Uma variação estatística de longo prazo das adversidades é determinada pelo estudo estatístico de ondas em longo prazo e ajustada por uma distribuição com base numa vida de projeto conhecida. Constam no trabalho resultados da variação da confiabilidade ao longo do tempo de um navio petroleiro. O caso de estudo foi simplificado para facilitar a obtenção de dados, com o objetivo de corroborar a metodologia desenvolvida.

Maintenance testing for the data encryption standard /

"Issued August 1980."

An empirical study of three stochastic approximation techniques applicable to sensitivity testing (U)

Photocopy.

Analysis of Melanoma Onset: Assessing Familial Aggregation by Using Estimating Equations and Fitting Variance Components via Bayesian Random Effects Models

We investigate whether relative contributions of genetic and shared environmental factors are associated with an increased risk in melanoma. Data from the Queensland Familial Melanoma Project comprising 15,907 subjects arising from 1912 families were analyzed to estimate the additive genetic, common and unique environmental contributions to variation in the age at onset of melanoma. Two complementary approaches for analyzing correlated time-to-onset family data were considered: the generalized estimating equations (GEE) method in which one can estimate relationship-specific dependence simultaneously with regression coefficients that describe the average population response to changing covariates; and a subject-specific Bayesian mixed model in which heterogeneity in regression parameters is explicitly modeled and the different components of variation may be estimated directly. The proportional hazards and Weibull models were utilized, as both produce natural frameworks for estimating relative risks while adjusting for simultaneous effects of other covariates. A simple Markov Chain Monte Carlo method for covariate imputation of missing data was used and the actual implementation of the Bayesian model was based on Gibbs sampling using the free ware package BUGS. In addition, we also used a Bayesian model to investigate the relative contribution of genetic and environmental effects on the expression of naevi and freckles, which are known risk factors for melanoma.

Using Bayesian neural networks to classify segmented images

We present results that compare the performance of neural networks trained with two Bayesian methods, (i) the Evidence Framework of MacKay (1992) and (ii) a Markov Chain Monte Carlo method due to Neal (1996) on a task of classifying segmented outdoor images. We also investigate the use of the Automatic Relevance Determination method for input feature selection.

Bayesian analysis of the scatterometer wind retrieval inverse problem:Some new approaches

The retrieval of wind vectors from satellite scatterometer observations is a non-linear inverse problem. A common approach to solving inverse problems is to adopt a Bayesian framework and to infer the posterior distribution of the parameters of interest given the observations by using a likelihood model relating the observations to the parameters, and a prior distribution over the parameters. We show how Gaussian process priors can be used efficiently with a variety of likelihood models, using local forward (observation) models and direct inverse models for the scatterometer. We present an enhanced Markov chain Monte Carlo method to sample from the resulting multimodal posterior distribution. We go on to show how the computational complexity of the inference can be controlled by using a sparse, sequential Bayes algorithm for estimation with Gaussian processes. This helps to overcome the most serious barrier to the use of probabilistic, Gaussian process methods in remote sensing inverse problems, which is the prohibitively large size of the data sets. We contrast the sampling results with the approximations that are found by using the sparse, sequential Bayes algorithm.

Neural Network Modelling with Input Uncertainty: Theory and Application

It is generally assumed when using Bayesian inference methods for neural networks that the input data contains no noise. For real-world (errors in variable) problems this is clearly an unsafe assumption. This paper presents a Bayesian neural network framework which accounts for input noise provided that a model of the noise process exists. In the limit where the noise process is small and symmetric it is shown, using the Laplace approximation, that this method adds an extra term to the usual Bayesian error bar which depends on the variance of the input noise process. Further, by treating the true (noiseless) input as a hidden variable, and sampling this jointly with the network’s weights, using a Markov chain Monte Carlo method, it is demonstrated that it is possible to infer the regression over the noiseless input. This leads to the possibility of training an accurate model of a system using less accurate, or more uncertain, data. This is demonstrated on both the, synthetic, noisy sine wave problem and a real problem of inferring the forward model for a satellite radar backscatter system used to predict sea surface wind vectors.

Studies on the generalisation of Gaussian processes and Bayesian neural networks

The assessment of the reliability of systems which learn from data is a key issue to investigate thoroughly before the actual application of information processing techniques to real-world problems. Over the recent years Gaussian processes and Bayesian neural networks have come to the fore and in this thesis their generalisation capabilities are analysed from theoretical and empirical perspectives. Upper and lower bounds on the learning curve of Gaussian processes are investigated in order to estimate the amount of data required to guarantee a certain level of generalisation performance. In this thesis we analyse the effects on the bounds and the learning curve induced by the smoothness of stochastic processes described by four different covariance functions. We also explain the early, linearly-decreasing behaviour of the curves and we investigate the asymptotic behaviour of the upper bounds. The effect of the noise and the characteristic lengthscale of the stochastic process on the tightness of the bounds are also discussed. The analysis is supported by several numerical simulations. The generalisation error of a Gaussian process is affected by the dimension of the input vector and may be decreased by input-variable reduction techniques. In conventional approaches to Gaussian process regression, the positive definite matrix estimating the distance between input points is often taken diagonal. In this thesis we show that a general distance matrix is able to estimate the effective dimensionality of the regression problem as well as to discover the linear transformation from the manifest variables to the hidden-feature space, with a significant reduction of the input dimension. Numerical simulations confirm the significant superiority of the general distance matrix with respect to the diagonal one.In the thesis we also present an empirical investigation of the generalisation errors of neural networks trained by two Bayesian algorithms, the Markov Chain Monte Carlo method and the evidence framework; the neural networks have been trained on the task of labelling segmented outdoor images.