No description available.

With the advent of autonomous systems and deep learning systems, safety pertaining to these systems has become a major concern. The existing failure analysis techniques are not enough to thoroughly analyze the safety in these systems. Moreover, because these systems are created to operate in various conditions, they are susceptible to unknown safety issues. Hence, we need mechanisms which can take into account the complexity of operational design domains, identify safety issues other than failures, and expose unknown safety issues. Moreover, existing safety analysis approaches require a lot of effort and time for analysis and do not consider machine learning (ML) safety. To address these limitations, in this dissertation, we discuss an artificial-intelligence driven model-based methodology that aids in identifying unknown safety issues and analyzing ML safety. Our methodology consists of 4 major tasks: 1) automated model generation, 2) automated analysis of component state transition model specification, 3) undesired states analysis, and 4) causal factor analysis. In our methodology we identify unknown safety issues by finding undesired combinations of components' states and environmental entities' states as well as causes resulting in these undesired combinations. In our methodology, we refer to the behaviors that occur because of undesired combinations as off-nominal behaviors (ONBs). To identify undesired combinations and ONBs that aid in exposing unknown safety issues with less effort and time we proposed various approaches for each of the task and performed corresponding empirical studies. We also discussed machine learning safety analysis from the perspective of machine learning engineers as well as system and software safety engineers. The results of studies conducted as part of our research shows that our proposed methodology helps in identifying unknown safety issues effectively. Our results also show that combinatorial methods are effective in reducing effort and time for analysis of off-nominal behaviors without overlooking any dependencies among components and environmental entities of a system. We also found that safety analysis of machine learning components is different from analysis of conventional software components and detail the aspects we need to consider for ML safety.

Ces travaux s'intéressent à l'estimation de la fiabilité des évacuateurs de crues vannés. Le comportement fiabilistes de ces systèmes hydrauliques dépend à la fois d'événements aléatoires discrets, mais aussi de l'évolution d'une variable déterministe continue : ce sont des systèmes dynamiques hybrides. Pour ces systèmes, l'événement redouté est réalisé lorsque le niveau de la retenue atteint un seuil de sûreté. La démarche de fiabilité dynamique proposée dans cette thèse vise à prendre en compte l’information temporelle de la modélisation à la synthèse d'indicateurs fiabilistes pour l'aide à la décision et développe deux contributions :1) L'élaboration d'une base de connaissances dédiée à la description des évacuateurs de crue en termes de fiabilité dynamique. Chaque classe de composants est décrite par un automate stochastique hybride dont les états sont les différentes phases de son fonctionnement. 2) Le suivi de la simulation de Monte Carlo et le traitement et l'analyse des "histoires" (séquence de tous les états activés et des dates d'activation) obtenues en simulation pour construire des indicateurs de fiabilité classique (probabilité d'occurrence de l'évènement redouté, identification des coupes équivalentes prépondérantes, ...). Des indicateurs de fiabilité dynamique basés sur la classification des histoires en fonction des dates de défaillance des composants concernés et sur l'estimation de l'importance dynamique sont aussi proposés / Hydraulic systems are hybrid dynamic systems whose evolution is a combination between discrete stochastic events on the one hand and continuous deterministic phenomena on the other hand. The undesired event is achieved when the dam level reaches a security threshold. In the frame of gated spillways dynamic reliability, the proposed methodology takes into account the temporal information during modeling and synthesis of reliability indicators for decision support.The first contribution of this work is the development of a knowledge base to describe a class of systems. Each component is described by a stochastic hybrid automaton whose states are the different working modes.The second contribution is Monte Carlo simulation monitoring and treatment results. A story is the sequence of all activated states and activation dates during the algorithm passage for a simulation. The analysis of results provides classical reliability indicators, such as the time evolution of the undesired event probability or the identification of predominant equivalent cuts. Our predictive approach is based on stories classification depending on components failure dates, then dynamic importance is assessed

The content of this diploma thesis is description of semi-automatic assembly line, description of design and realization of electroprojection including electrical switchboard and control panel, together with description of design and implementation of control system including communications and user interface. Part of the thesis is also the analysis and solution of safety risks using a safety controller and overall verification and evaluation of the functionality of the whole system on described line.

This thesis deals with the analysis and design changes to a specific company information system under real conditions, and is therefore more practically oriented. The information system is studied both from the perspective of the end user, which is the focus, and in terms of operation and implementation of program solutions. The information system is studied as a whole in a broader context, not only in terms of applications (software), but also in terms of the flow of information, storage and organization of data, users, security, hardware, etc. It should serve to the company as a guide for the elimination of inefficient and / or risk areas, thereby helping to improve work productivity and employee satisfaction.

Les travaux présentés dans ce manuscrit visent à développer des modèles de coût/performances pour améliorer les décisions de maintenance sur les infrastructures ferroviaires exploitées dans un environnement de plus en plus contraint: trafic accru, détérioration accélérée, temps de maintenance réduits. Les modèles de maintenance proposés sont construits à base de réseaux de Petri colorés ; ils sont animés par simulation de Monte Carlo pour estimer les performances (en termes de coût et de disponibilité) des politiques de maintenance considérées. Ils sont développés aux niveaux "composant" et "réseau", et plusieurs problèmes de maintenance différents sont étudiés. Au niveau "composant" (rail), des politiques de maintenance mettant en jeu différents niveaux d'information de surveillance sont comparées pour montrer l'intérêt de surveiller la détérioration graduelle du composant. L'effet de l'existence d'un délai de maintenance est également étudié pour les politiques conditionnelle et périodique. Au niveau système (ligne), une maintenance mettant en jeu différents types d'inspections complémentaires (automatique ou visuelle) est d'abord étudiée. On s'intéresse ensuite au cas de figure où l'évolution de la détérioration dépend du mode d'utilisation et de la charge de la voie : le problème de maintenance étudié vise alors à définir un réglage optimal des paramètres d'exploitation de la voie (vitesse limite) et de maintenance (délai d'intervention) / The aim of this thesis research work is to propose maintenance models for railways infrastructures that can help to make better maintenance decisions in the more constrained environment that the railway industry has to face, e.g. increased traffic loads, faster deterioration, longer maintenance planning procedures, shorter maintenance times. The proposed maintenance models are built using Coloured Petri nets; they are animated through Monte Carlo simulations to estimate the performance of the considered maintenance policies in terms of cost and availability. The maintenance models are developed both at the component and network levels, and several different maintenance problems are considered. At the rail component level, maintenance policies with different level of monitoring information (level of gradual deterioration vs binary working state) are compared to show the benefits of gathering monitoring information on the deterioration level. The effect of preventive maintenance delays is also investigated for both condition-based inspection policies and periodic inspection policies on a gradually deteriorating component. At the line level, a maintenance policy based on a two-level inspection procedure is first investigated. Then, considering the case when the deterioration process depends on the operation modes (normal vs limited speed), a maintenance optimization problem is solved to determine an optimal tuning of the repair delay and speed restriction

Each year people are injured and even die in crowd related accidents, often during planned events. Recent studies have emphasized the need for using a systems approach to study these events. In this study the systems approach of resilience theory is combined with the crowd psychology-models Extended Social Identity Model and the Aggravation and Mitigation Model to examine event safety at a music festival, a domain previously largely unexplored by these perspectives. By using an ethnographic approach as well as interviewing visitors the study set out to answer questions about when and how the safety organization adjusted itself under conditions relating to crowds. Another goal was to study the social identity of the visitors as well as the interaction between the safety organization and the visitors at the festival to explain the presence or absence of crowd conflict. Using thematic analysis several situations were identified where the safety organization adjusted itself, as well as the strategies that the organization used in these different circumstances. It was also concluded that the absence of crowd conflict could best be explained by three factors. First of all, no history of crowd conflict existed between the safety organization and the visitors, secondly, there were no groups present with the goal of creating conflict, and thirdly, the social processes taking place between the safety organization and the visitors were all mitigating in nature. The mitigating nature of the social processes was partly attributable to the strategies identified for adjusting to crowd conditions.

Mit der steigenden Nachfrage nach Innovationen spielt Software in verschiedenenWirtschaftsbereichen eine wichtige Rolle, wie z.B. in der Automobilindustrie, bei intelligenten Systemen als auch bei Kommunikationssystemen. Daher ist die Qualität für die Softwareentwicklung von großer Bedeutung. Allerdings ändern sich die probabilistische Modelle (die Qualitätsbewertungsmodelle) angesichts der dynamischen Natur moderner Softwaresysteme. Dies führt dazu, dass ihre Übergangswahrscheinlichkeiten im Laufe der Zeit schwanken, welches zu erheblichen Problemen führt. Dahingehend werden probabilistische Modelle im Hinblick auf ihre Laufzeit kontinuierlich aktualisiert. Eine fortdauernde Neubewertung komplexer Wahrscheinlichkeitsmodelle ist jedoch teuer. In letzter Zeit haben sich inkrementelle Ansätze als vielversprechend für die Verifikation von adaptiven Systemen erwiesen. Trotzdem wurden bei der Bewertung struktureller Änderungen im Modell noch keine wesentlichen Verbesserungen erzielt. Wahrscheinlichkeitssysteme werden als Automaten modelliert, wie bei Markov-Modellen. Solche Modelle können in Matrixform dargestellt werden, um die Gleichungen basierend auf Zuständen und Übergangswahrscheinlichkeiten zu lösen. Laufzeitmodelle wie Matrizen sind nicht signifikant, um die Auswirkungen von Modellveränderungen erkennen zu können. In dieser Arbeit wird ein Framework unter Verwendung stochastischer Bäume mit regulären Ausdrücken entwickelt, welches modular aufgebaut ist und eine aktionshaltige sowie probabilistische Logik im Kontext der Modellprüfung aufweist. Ein solches modulares Framework ermöglicht dem Menschen die Entwicklung der Änderungsoperationen für die inkrementelle Berechnung lokaler Änderungen, die im Modell auftreten können. Darüber hinaus werden probabilistische Änderungsmuster beschrieben, um eine effiziente inkrementelle Verifizierung, unter Verwendung von Bäumen mit regulären Ausdrücken, anwenden zu können. Durch die Bewertung der Ergebnisse wird der Vorgang abgeschlossen. / Software plays an innovative role in many different domains, such as car industry, autonomous and smart systems, and communication. Hence, the quality of the software is of utmost importance and needs to be properly addressed during software evolution. Several approaches have been developed to evaluate systems’ quality attributes, such as reliability, safety, and performance of software. Due to the dynamic nature of modern software systems, probabilistic models representing the quality of the software and their transition probabilities change over time and fluctuate, leading to a significant problem that needs to be solved to obtain correct evaluation results of quantitative properties. Probabilistic models need to be continually updated at run-time to solve this issue. However, continuous re-evaluation of complex probabilistic models is expensive. Recently, incremental approaches have been found to be promising for the verification of evolving and self-adaptive systems. Nevertheless, substantial improvements have not yet been achieved for evaluating structural changes in the model. Probabilistic systems are usually represented in a matrix form to solve the equations based on states and transition probabilities. On the other side, evolutionary changes can create various effects on theese models and force them to re-verify the whole system. Run-time models, such as matrices or graph representations, lack the expressiveness to identify the change effect on the model. In this thesis, we develop a framework using stochastic regular expression trees, which are modular, with action-based probabilistic logic in the model checking context. Such a modular framework enables us to develop change operations for the incremental computation of local changes that can occur in the model. Furthermore, we describe probabilistic change patterns to apply efficient incremental quantitative verification using stochastic regular expression trees and evaluate our results.