Dans la robotique mobile, le réseau de communication est un composant important du système global pour que le système accomplisse sa mission. Dans un tel type de système, appelé un système commandé en réseau sans fil (SCR sans fil ou WNCS), l’intégration du réseau sans fil dans la boucle de commande introduit des problèmes qui ont un impact sur la performance et la stabilité i.e, sur la qualité de commande (QoC). Cette QoC dépend alors de la qualité de service (QoS) et la performance du système va donc dépendre des paramètres de la QoS. C’est ainsi que l’étude de l’influence des défauts du réseau sans fil sur la QoC est cruciale. Le WNCS est un système temps réel qui a besoin d’un certain niveau de QoS pour une bonne performance. Cependant, la nature probabiliste du protocole de communication CSMA/CA utilisé dans la plupart des technologies sans fil ne garantit pas les contraintes temps réel. Il faut alors une méthode probabiliste pour analyser et définir les exigences de l’application en termes de QoS, c’est-à-dire en termes de délai, de gigue, de débit, et de perte de paquets. Une première contribution de cette thèse consiste à étudier les performances et la fiabilité d’un réseau sans fil IEEE 802.11 pour des WNCSs qui partagent le même réseau et le même serveur de commandes en développant un modèle stochastique. Ce modèle est une chaîne de Markov qui modélise la méthode d’accès au canal de communication. Ce modèle a servi pour définir les paramètres de la QoS qui peuvent garantir une bonne QoC. Nous appliquons notre approche à un robot mobile commandé par une station distante. Le robot mobile a pour mission d’atteindre une cible en évitant les obstacles. Pour garantir l’accomplissement de cette mission, une méthode de diagnostic probabiliste est primordiale puisque le comportement du système n’est pas déterministe. La deuxième contribution a été d’établir la méthode probabiliste qui sert à surveiller le bon déroulement de la mission et l’état du robot. C’est un réseau bayésien (RB) modulaire qui modélise les relations de dépendance cause-à-effet entre les défaillances qui ont un impact sur la QoC du système. La dégradation de la QoC peut être due soit à un problème lié à l’état interne du robot, soit à un problème lié à la QoS, soit à un problème lié au contrôleur lui-même. Les résultats du modèle markovien sont utilisés dans le RB modulaire pour définir l'espace d'état de ses variables (étude qualitative) et pour définir les probabilités conditionnelles de l'état de la QoS (étude quantitative). Le RB permet d’éviter la dégradation de la QoC en prenant la bonne décision qui assure la continuité de la mission. En effet, dans une approche de co-design, quand le RB détecte une dégradation de la QoC due à une mauvaise QoS, la station envoie un ordre au robot pour qu'il change son mode de fonctionnement ou qu'il commute sur un autre contrôleur débarqué. Notre hypothèse est que l’architecture de diagnostic est différente en fonction des modes de fonctionnement : nous optons pour un RB plus global et partagé lorsque le robot est connecté à la station et pour RB interne au robot lorsqu’il est autonome. La commutation d’un mode de fonctionnement débarqué à un mode embarqué implique la mise à jour du RB. Un autre apport de cette thèse est la définition d’une stratégie de commutation entre les modes de diagnostic : commutation d’un RB distribué à un RB monolithique embarqué quand le réseau de communication ne fait plus partie de l'architecture du système et vice-versa. Les résultats d’inférence et de scénario de diagnostic ont montré la pertinence de l’utilisation des RBs distribués modulaires. Ils ont aussi montré la capacité du RB développé à détecter la dégradation de la QoC et de la QoS et à superviser l’état du robot. L’aspect modulaire du RB a permis de faciliter la reconfiguration de l’outil de diagnostic selon l’architecture de commande ou de communication adaptée (RB distribué ou RB monolithique embarqué). / In mobile robotics systems, the communication network is an important component of the overall system, it enables the system to accomplish its mission. Such a system is called Wireless Networked Control System WNCS where the integration of the wireless network into the control loop introduces problems that impact its performance and stability i.e, its quality of control (QoC). This QoC depends on the quality of service (QoS) therefore, the performance of the system depends on the parameters of the QoS. The study of the influence of wireless network defects on the QoC is crucial. WNCS is considered as a real-time system that requires a certain level of QoS for good performance. However, the probabilistic behavior of the CSMA / CA communication protocol used in most wireless technologies does not guarantee real-time constraints. A probabilistic method is then needed to analyze and define the application requirements in terms of QoS: delay, jitter, rate, packet loss. A first contribution of this thesis is to study the performance and reliability of an IEEE 802.11 wireless network for WNCSs that share the same network and the same control server by developing a stochastic model. This model is a Markov chain that models the access procedure to the communication channel. This model is used to define the QoS parameters that can guarantee the good QoC. In this thesis, we apply our approach to a mobile robot controlled by a remote station. The mobile robot aims to reach a target by avoiding obstacles, a classic example of mobile robotics applications. To ensure that its mission is accomplished, a probabilistic diagnostic method is essential because the system behavior is not deterministic. The second contribution of this thesis is to establish the probabilistic method used to monitor the robot mission and state. It is a modular Bayesian network BN that models cause-and-effect dependency relationships between failures that have an impact on the system QoC. The QoC degradation may be due either to a problem related to the internal state of the robot, a QoS problem or a controller problem. The results of the Markov model analysis are used in the modular BN to define its variables states (qualitative study) and to define the conditional probabilities of the QoS (quantitative study). It is an approach that permits to avoid the QoC degradation by making the right decision that ensures the continuity of the mission. In a co-design approach, when the BN detects a degradation of the QoC due to a bad QoS, the station sends an order to the robot to change its operation mode or to switch to another distant controller. Our hypothesis is that the diagnostic architecture depends on the operation mode. A distributed BN is used when the robot is connected to the station and a monolithic embedded BN when it is autonomous. Switching from a distributed controller to an on-board one involves updating the developed BN. Another contribution of this thesis consists in defining a switching strategy between the diagnostic modes: switching from a distributed BN to an on-board monolithic BN when the communication network takes no longer part of the system architecture and vice versa -versa. The inference and diagnostic scenarii results show the relevance of using distributed modular BNs. They also prove the ability of the developed BN to detect the degradation of QoC and QoS and to supervise the state of the robot. The modular structure of the BN facilitates the reconfiguration of the diagnostic policy according to the adapted control and communication architecture (distributed BN or on-board monolithic RB).

Grandes avanços têm sido alcançados com as técnicas de análise de segurança dos sistemas essenciais de navegação e performance das aeronaves resultando na diminuição das taxas de acidentes ao longo dos últimos anos. O Relatório de Segurança de 2017 da EASA (European Agency Safety Aviation) apresenta um relevante aumento do número de acidentes não fatais. Este resultado positivo leva ao aumento das evacuações de emergência. O Relatório de Segurança de 2016 da IATA (International Air Transport Association) mostra que em 35% dos acidentes com sobreviventes em Jatos e 55% dos acidentes com sobreviventes em turbo hélice ocorreram com evacuação de emergência. Diante deste cenário, a confiabilidade humana torna-se relevante na interface destes passageiros com o projeto de segurança da cabine durante o procedimento de evacuação de emergência. Para avaliar as características e a contribuição desta interface no sucesso do procedimento de evacuação, é proposta uma metodologia para a análise da interação humana com este sistema estabelecendo um diagrama causal genérico com o objetivo de estudar o mecanismo do erro humano nesta interface. A metodologia proposta utiliza a abordagem das Redes Bayesianas apoiada pela lógica Fuzzy para modelar os Fatores de Desempenho Humano e para verificar, através da diagnose e inferência causal, quais fatores mais influenciam o desempenho humano na execução das tarefas neste ambiente de emergência. Esta pesquisa apresenta uma aplicação da metodologia proposta para analisar as tarefas do ensaio de evacuação de emergência de uma aeronave, focando na quantificação do erro humano na interface com o projeto de segurança da cabine da aeronave. Os resultados da aplicação identificaram o fator situacional: cartão de segurança, marcas na asa e escorregadores, e os fatores individuais: conhecimento e habilidades: interpretação e percepção como aqueles que mais influenciaram no teste do procedimento de evacuação de emergência de uma aeronave. / Great advances have been achieved with the safety assessment techniques of essential aircraft navigation and performance systems due to decreasing of fatal accident rates in recent years. The EASA Annual Safety Report 2017 (European Agency Safety Aviation) presents a relevant increase of non-fatal accidents. This positive results leads to increasing of emergency evacuation. The IATA Safety Report 2016 (International Air Transport Association) presents that 35% of survival accidents with Jet and 55% of survival accidents with Turboprop occurred with emergency evacuation. In view of this scenario, human reliability becomes relevant in the interface of these passengers with the cabin safety design during emergency evacuation procedure. To evaluate this interface features, and the contribution of this interface in the success of evacuation procedure, it is proposed a method for analyzing the human interaction within the system, to establish a generic causal framework aiming at the study of the human error mechanism. The proposed methodology uses the Bayesian Networks approach supported by Fuzzy logic for modelling Human Performance Factors and for verifying, through diagnosis and causal inference, which factors most influence human performance in the execution of tasks in this emergency environment. This research presents an application of this approach to analyze the tasks of the emergency evacuation testing from an aircraft, focusing on the quantification of human error in the interface with aircraft cabin safety design. The results of application has identified the situational factor: safety card, marks on the wing and escape slides, and the individual factors: knowledge and abilities: interpretation and perception as one those most of influenced the emergency evacuation test procedure from an aircraft.

Power disspiation is a growing concern in VLSI circuits. In this work we model the data dependence of power dissipation by learning an input model which we use for estimation of both switching activity and crosstalk for every node in the circuit. We use Bayesian networks to effectively model the spatio-temporal dependence in the inputs and we use the probabilistic graphical model to learn the structure of the dependency in the inputs. The learned structure is representative of the input model. Since we learn a causal model, we can use a larger number of independencies which guarantees a minimal structure. The Bayesian network is converted into a moral graph, which is then triangulated. The junction tree is formed with its nodes representing the cliques. Then we use logic sampling on the junction tree and the sample required is really low. Experimental results with ISCAS '85 benchmark circuits show that we have achieved a very high compaction ratio with average error less than 2%. As HSPICE was used the results are the most accurate in terms of delay consideration. The results can further be used to predict the crosstalk between two neighboring nodes. This prediction helps in designing the circuit to avoid these problems.

Power dissipation in a VLSI circuit poses a serious challenge in present and future VLSI design. A switching model for the data dependent behavior of the transistors is essential to model dynamic, load-dependent active power and also leakage power in active mode - the two components of power in a VLSI circuit. A probabilistic Bayesian Network based switching model can explicitly model all spatio-temporal dependency relationships in a combinational circuit, resulting in zero-error estimates. However, the space-time requirements of exact estimation schemes, based on this model, increase with circuit complexity [5, 24]. This work explores a non-simulative, importance sampling based, probabilistic estimation strategy that scales well with circuit complexity. It has the any-time aspect of simulation and the input pattern independence of probabilistic models. Experimental results with ISCAS'85 benchmark shows a significant savings in time (nearly 3 times) and significant reduction in maximum error (nearly 6 times) especially for large benchmark circuits compared to the existing state of the art technique (Approximate Cascaded Bayesian Network) which is partition based. We also present a novel probabilistic method that is not dependent on the pre-specification of input-statistics or the availability of input-traces, to identify nodes that are likely to be leaky even in the active zone. This work emphasizes on stochastic data dependency and characterization of the input space, targeting data-dependent leakage power. The central theme of this work lies in obtaining the posterior input data distribution, conditioned on the leakage at an individual signal. We propose a minimal, causal, graphical probabilistic model (Bayesian Belief Network) for computing the posterior, based on probabilistic propagation flow against the causal direction, i.e. towards the input. We also provide two entropy-based measures to characterize the amount of uncertainties in the posterior input space as an indicator of the likelihood of the leakage of a signal. Results on ISCAS'85 benchmark shows that conclusive judgments can be made on many nodes without any prior knowledge about the input space.

Management of data imprecision has become increasingly important, especially with the advance of technology enabling applications to collect and store huge amount data from multiple sources. Data collected in such applications involve a large number of variables and various types of data imperfections. These data, when used in knowledge discovery applications, require the following: 1) computationally efficient algorithms that works faster with limited resources, 2) an effective methodology for modeling data imperfections and 3) procedures for enabling knowledge discovery and quantifying and propagating partial or incomplete knowledge throughout the decision-making process. Bayesian Networks (BNs) provide a convenient framework for modeling these applications probabilistically enabling a compact representation of the joint probability distribution involving large numbers of variables. BNs also form the foundation for a number of computationally efficient algorithms for making inferences. The underlying probabilistic approach however is not sufficiently capable of handling the wider range of data imperfections that may appear in many new applications (e.g., medical data). Dempster-Shafer theory on the other hand provides a strong framework for modeling a broader range of data imperfections. However, it must overcome the challenge of a potentially enormous computational burden. In this dissertation, we introduce the joint Dirichlet BoE, a certain mass assignment in the DS theoretic framework, that simplifies the computational complexity while enabling one to model many common types of data imperfections. We first use this Dirichlet BoE model to enhance the performance of the EM algorithm used in learning BN parameters from data with missing values. To form a framework of reasoning with the Dirichlet BoE, the DS theoretic notions of conditionals, independence and conditional independence are revisited. These notions are then used to develop the DS-BN, a BN-like graphical model in the DS theoretic framework, that enables a compact representation of the joint Dirichlet BoE. We also show how one may use the DS-BN in different types of reasoning tasks. A local message passing scheme is developed for efficient propagation of evidence in the DS-BN. We also extend the use of the joint Dirichlet BoE to Markov models and hidden Markov models to address the uncertainty arising due to inadequate training data. Finally, we present the results of various experiments carried out on synthetically generated data sets as well as data sets from medical applications.

No description available.

<p>Many biological processes are effectively modeled as networks, but a frequent assumption is that these networks do not change during data collection. However, that assumption does not hold for many phenomena, such as neural growth during learning or changes in genetic regulation during cell differentiation. Approaches are needed that explicitly model networks as they change in time and that characterize the nature of those changes.</p><p>In this work, we develop a new class of graphical models in which the conditional dependence structure of the underlying data-generation process is permitted to change over time. We first present the model, explain how to derive it from Bayesian networks, and develop an efficient MCMC sampling algorithm that easily generalizes under varying levels of uncertainty about the data generation process. We then characterize the nature of evolving networks in several biological datasets.</p><p>We initially focus on learning how neural information flow networks change in songbirds with implanted electrodes. We characterize how they change in response to different sound stimuli and during the process of habituation. We continue to explore the neurobiology of songbirds by identifying changes in neural information flow in another habituation experiment using fMRI data. Finally, we briefly examine evolving genetic regulatory networks involved in Drosophila muscle differentiation during development.</p><p>We conclude by suggesting new experimental directions and statistical extensions to the model for predicting novel neural flow results.</p> / Dissertation

Taiwan, as an island country, should place future aspiration on the usages of ocean energy and marine resources, such as offshore wind power and deep ocean water. The sound development of marine services relies on a strong industry of maritime engineering. The perilous marine environment has posed the highest risk for all maritime civil engineering activities. It is therefore imperative to restrain the risk associated with current maritime work, other than just engineering technique itself. By doing so, the quality of maritime work can be assured, and as the improvement of overall engineering capability, Taiwan can compete worldwide in the maritime engineering industry. Maritime works have developed their own standard construction procedures. To mitigate risk of maritime works depend mainly on the domain experts¡¦ experience and know-how. However, problems appear when less experienced experts are available, or qualitative experience exists in a narrative form. It is therefore important to structure clearly an engineering risk factor relation, and quantify and control these risk factors. The proposed study will first collect and review related literatures, and then interview an expert from the designate maritime service company to establish the risk factors associated with offshore piping. Eventually a complete Bayesian network (BN) was formulated based on the cause-effect diagram, using Interpretive Structural Modeling (ISM), and experts¡¦ experience was transformed into a set of prior and conditional probability to be embedded in the BN. The BN can clearly show that certain earlier operational factors affect final operational process deeply. Besides, the backward reasoning using the BN is possible to identify the factors causing a project failure.

Web service selection problem focuses on how to choose component Web services to satisfy user¡¦s non-functional (or QoS) need, and it has been extensively studied in the past. The QoS measures include reliability, response time, and execution cost. However, in some applications, execution result, as demonstrated on some output values, matters, and this is seldom addressed by previous researches. In our work, we proposed an approach to guide the WS selection with the goal to meet user¡¦s preferences on the composition execution outcome. In addition, we consider the partner relationship between Web services. Some partner Web services may produce more desired execution result, such as better quality or a discount, than others. In our approach, we use Bayesian Network to guide Web services selection. Specifically, we propose two Bayesian Network-based methods: Partner-first Bayesian Network and Probability-first Bayesian Network. Both methods rank Web services by considering user¡¦s preference, user¡¦s input variables, and the past execution results of Web services. The experiment result shows that the proposed Bayesian Network methods perform better than the other more straightforward methods.

Web services can be broadly classified into two types, namely effect providing (EP) services and data providing (DP) services. In this work, we address DP service composition problem that intends to satisfy user preference specified at the instance level, namely the expected occurrence. We first use the query rewriting method to identify a composition of service types that satisfies user¡¦s requirement and employ Bayesian Network model to express the causal relationship between exchange variables of DP service types. Service selection is then conducted by computing the posterior probability in the Bayesian Network. We conduct experiments to show that our proposed Bayesian Network-based method outperforms the other baseline methods in terms of execution success rate and data quality. It also has reasonable execution time.