Self-healing systems promise operating cost reductions in large-scale computing environments through the automated detection of, and recovery from, faults. However, at present there appears to be little known empirical evidence comparing the different approaches, or demonstrations that such implementations reduce costs. This thesis compares previous and current self-healing approaches before demonstrating a new, unsupervised approach that combines artificial neural networks with performance tests to perform fault identification in an automated fashion, i.e. the correct and accurate determination of which computer features are associated with a given performance test failure. Several key contributions are made in the course of this research including an analysis of the different types of self-healing approaches based on their contextual use, a baseline for future comparisons between self-healing frameworks that use artificial neural networks, and a successful, automated fault identification in cloud infrastructure, and more specifically virtual machines. This approach uses three established machine learning techniques: Naïve Bayes, Baum-Welch, and Contrastive Divergence Learning. The latter demonstrates minimisation of human-interaction beyond previous implementations by producing a list in decreasing order of likelihood of potential root causes (i.e. fault hypotheses) which brings the state of the art one step closer toward fully self-healing systems. This thesis also examines the impact of that different types of faults have on their respective identification. This helps to understand the validity of the data being presented, and how the field is progressing, whilst examining the differences in impact to identification between emulated thread crashes and errant user changes – a contribution believed to be unique to this research. Lastly, future research avenues and conclusions in automated fault identification are described along with lessons learned throughout this endeavor. This includes the progression of artificial neural networks, how learning algorithms are being developed and understood, and possibilities for automatically generating feature locality data.

Les logiciels s'orientent de plus en plus vers des architectures évolutives, capables de s'adapter facilement aux changements et d'intégrer de nouvelles fonctionnalités. Ceci est important pour plusieurs classes d'applications qui ont besoin d‘évoluer sans que cela implique d'interrompre leur exécution. Des plateformes dynamiques à composants autorisent ce type d'évolution à l'exécution, en permettant aux composants d'être chargés et exécutés sans requérir le redémarrage complet de l'application en service. Toutefois, la flexibilité d'un tel mécanisme introduit de nouveaux défis qui exigent de gérer les possibles erreurs dues à des incohérences dans le processus de mise à jour, ou en raison du comportement défectueux de composants survenant pendant l'exécution de l'application. Des composants tiers dont l'origine ou la qualité sont inconnus peuvent être considérées à priori comme peu fiables, car ils peuvent potentiellement introduire des défauts d'applications lorsqu'il est combiné avec d'autres composants. Nous sommes intéressés à la réduction de l'impact de ces composants considérés comme non fiables et qui sont susceptibles de compromettre la fiabilité de l'application en cours d'exécution. Cette thèse porte sur l'application de techniques pour améliorer la fiabilité des applications dynamiques à composants. Pour cela, nous proposons l'utilisation des frontières d'isolation pouvant fournir du contingentement de fautes. Le composant ainsi isolé ne perturbe pas le reste de l'application quand il est défaillant. Une telle approche peut être vu sous trois perspectives présentées: (i) l'isolement des composants dynamiques, régi par une politique d'exécution reconfigurable, (ii) l'autoréparation de conteneurs d‘isolement, et (iii) l'utilisation des aspects pour séparer les préoccupations de fiabilité à partir du code fonctionnel. / Software is moving towards evolutionary architectures that are able to easily accommodate changes and integrate new functionality. This is important in a wide range of applications, from plugin-based end user applications to critical applications with high availability requirements. Dynamic component-based platforms allow software to evolve at runtime, by allowing components to be loaded, and executed without forcing applications to be restarted. However, the flexibility of such mechanism demands applications to cope with errors due to inconsistencies in the update process, or due to faulty behavior from components introduced during execution. This is mainly true when dealing with third-party components, making it harder to predict the impacts (e.g., runtime incompatibilities, application crashes) and to maintain application dependability when integrating such third-party code into the application. Components whose origin or quality attributes are unknown could be considered as untrustworthy since they can potentially introduce faults to applications when combined with other components, even if unintentionally. The quality of components is harder to evaluate when components are combined together, especially if it happens on-the-fly. We are interested in reducing the impact that can be brought by untrustworthy components deployed at runtime and that would potentially compromise application dependability. This thesis focuses on applying techniques for moving a step forward towards dependable dynamic component-based applications by addressing different dependability attributes namely reliability, maintainability and availability. We propose the utilization of strong component isolation boundaries, by providing a fault-contained environment for separately running untrustworthy components. Our solution combines three approaches: (i) the dynamic isolation of components, governed by a runtime reconfigurable policy; (ii) a self-healing component isolation container; and (iii) the usage of aspects for separating dependability concerns from functional code.

L’alliage d’aluminium 2024 est très répandu dans l’industrie aéronautique et automobile. Le processus de corrosion peut entrainer des dommages irréversibles pouvant engendrer des issues fatales dans le domaine aéronautique. Ainsi plusieurs couches de revêtements sont déposées à la surface du métal à protéger pour prévenir le processus de corrosion. Depuis le début du 20ième siècle, le chrome hexavalent (CrVI) a été largement utilisé comme inhibiteur de corrosion dans les différentes couches du revêtement (prétraitement, primer et top-coat). La toxicité de ce composé envers l’homme et l’environnement a entrainé son interdiction et donc son remplacement comme agent de protection. Le confinement d’agent anticorrosif dans des nano-conteneurs a ainsi été reporté puisqu’un effet auto-réparant, en relargant sur demande, peut-être apporté : l’inhibiteur de corrosion agit quand le dommage apparait. Cette étude est focalisée sur l’utilisation de matériaux de type Hydroxydes Doubles Lamellaires (HDL) comme réservoir d’inhibiteurs de corrosion en raison leur propriété d’échange. Dans ce travail, plusieurs molécules ont été étudiées comme potentiel inhibiteur de corrosion. Celles-ci ont été tout d’abord caractérisées par DC-polarisation afin de déterminer la nature de leur comportement inhibiteur (anodique, cathodique ou les deux). Ensuite, ces agents anticorrosifs ont été intercalés dans des matrices HDL et leur capacité de relargage ainsi que leur comportement face au processus de corrosion ont été étudiés. Les particules HDL ainsi obtenues ont été dispersées dans la formulation d’un revêtement primaire et déposé directement sur l’alliage aluminium 2024. La résistance à la corrosion a été suivie par spectroscopie d’impédance complexe. Les propriétés d’auto-protection de notre revêtement (relargage d’agent anticorrosive encas de dommage) et leur propriété barrière, apportée par la morphologie lamellaire des nano-conteneurs, entrainant une diminution de la perméabilité aux espèces agressives (ex. eau, O2, électrolyte) responsable de l’apparition de cloques sur les revêtements, ont ainsi été caractérisées. / Aluminum alloy 2024 is widely used in aircraft and automotive industry. Corrosion processes can provide irreversible damage on the metal substrate which could have a tragic issue in the aircraft domain. Thus, several coating layers have been applied on the metal substrate to prevent corrosion process. Since the beginning of the 20th century, hexavalent chromate compounds have been extensively used as corrosion inhibitor agents for paint, primer and conversion coating. The toxicity for human health and environment has led to replace such compounds. The literature has reported different possibilities to replace such unfriendly compounds. Moreover, the entrapment of corrosion inhibitors in nanocontainer provides a self-healing effect by releasing, on demand, the active species when damage occurs. This study focuses on Layered Double Hydroxide (LDH) material as reservoir due to its exchange properties. This study has characterized several potential corrosion inhibitor molecules by DC-Polarization to determine the nature of the inhibitor compound (i.e. anodic, cathodic or both of them). Further, active anticorrosive species have been intercalated into LDH framework. Then, the release of inhibitor agents and their subsequent behaviour toward corrosion inhibition have been evaluated. Modified LDH materials have been further dispersed in the primer coating formulation and applied on aluminum alloys 2024 substrate. Corrosion inhibition has been followed by electrochemical impedance spectroscopy experiments on scratched and unscratched panel which have permitted to evaluate the self-healing property of these modified LDH materials when damage occurs and the barrier property provided by the lamellar morphology of the inorganic reservoir that is found to decrease the permeation by enhancing the tortuosity of the coating layer towards aggressive species (i.e. water, O2, electrolyte) responsible of the blistering phenomenon.

The advent of deep submicron technology coupled with ever increasing demands from the customer for more functionality on a compact silicon real estate has led to a proliferation of highly complex integrated RF system-on-chip (SoC) and system-on-insulator (SoI) solutions. The use of scaled CMOS technologies for high frequency wireless applications is posing daunting technological challenges both in design and manufacturing test. To ensure market success, manufacturers need to ensure the quality of these advanced RF devices by subjecting them to a conventional set of production test routines that are both time consuming and expensive. Typically the devices are tested for parametric specifications such as gain, linearity metrics, quadrature mismatches, phase noise, noise figure (NF) and end-to-end system level specifications such as EVM (error vector magnitude), BER (bit-error-rate) etc. Due to the reduced visibility imposed by high levels of integration, testing for parametric specifications are becoming more and more complex. To offset the yield loss resulting from process variability effects and reliability issues in RF circuits, the use of self-healing/self-tuning mechanisms will be imperative. Such self-healing is typically implemented as a test/self-test and self-tune procedure and is applied post-manufacture. To enable this, simple test routines that can accurately diagnose complex performance parameters of the RF circuits need to be developed first. After diagnosing the performance of a complex RF system appropriate compensation techniques need to be developed to increase or restore the system performance. Moreover, the test, diagnosis and compensation approach should be low-cost with minimal hardware and software overhead to ensure that the final product is economically viable for the manufacturer. The main components of the thesis are as follows: 1) Low-cost specification testing of advanced radio frequency front-ends: Methodologies are developed to address the issue of test cost and test time associated with conventional production testing of advanced RF front-ends. The developed methodologies are amenable for performing self healing of RF SoCs. Test generation algorithms are developed to perform alternate test stimulus generation that includes the artifacts of test signal path such as response capture accuracy, load-board DfT etc. A novel cross loop-back methodology is developed to perform low cost system level specification testing of multi-band RF transceivers. A novel low-cost EVM testing approach is developed for production testing of wireless 802.11 OFDM front-ends. A signal transformation based model extraction technique is developed to compute multiple RF system level specifications of wireless front-ends from a single data capture. The developed techniques are low-cost and facilitate a reduction in the overall contribution of test cost towards the manufacturing cost of advanced wireless products. 2)Analog tuning methodologies for compensating wireless RF front ends: Methodologies for performing low-cost self tuning of multiple impairments of wireless RF devices are developed. This research considers for the first time, multiple analog tuning parameters of a complete RF transceiver system (transmitter and receiver) for tuning purposes. The developed techniques are demonstrated on hardware components and behavioral models to improve the overall yield of integrated RF SoCs.

This thesis proposes a multifaceted production test and post-manufacture yield enhancement framework for RF systems. This framework uses low-cost test and post-manufacture calibration/tuning techniques. Since the test cost and the yield of the RF circuits/sub-system directly contribute to the manufacturing cost of RF systems, the proposed framework minimizes overall RF systems' manufacturing cost by taking two approaches. In the first approach, low-cost testing methodologies are proposed for RF amplifiers and integrated RF substrates with an embedded RF passive filter and interconnect. Techniques are developed to test RF circuits by the analysis of low-frequency signal of the order of few MHz and without using any external RF test-stimulus. Oscillation principles are used to enable testing of RF circuits without any external test-stimulus. In the second approach, to increase the yield of the RF circuits for parametric defects, RF circuits are tuned to compensate for a performance loss during production test using on-board or on-chip resources. This approach includes a diagnosis algorithm to identify faulty circuits within the system, and performs a compensation process that adjusts tunable components to enhance the performance of the RF circuits. In the proposed yield improvement methodologies, the external test stimulus is not required because the stimulus is generated by the RF circuit itself with the help of additional circuitry and faulty circuits are detected using low-cost test methods developed in this research. As a result, the proposed research enables low-cost testing and self-healing of RF systems.

In den durchgeführten Versuchsreihen wurden die Wasserabsorption sowie die Gas- und Wasserpermeabilität sowohl an gerissenen als auch rissfreien Prüfkörpern aus textilbewehrtem Beton (TRC) untersucht. Eine deutliche Steigerung der Wasseraufnahme bei Proben mit unbeschichteten Textilien konnte mit Zunahme der Garnfeinheit beobachtet werden. Bei den gerissenen Proben besteht eine ausgeprägte Abhängigkeit der Transportraten von Flüssigkeiten und Gasen zu den relevanten Risscharakteristika (kumulative Risslänge, Rissbreite). Weiterhin sind Selbstheilungseffekte von feinen Rissen infolge einer zyklischen Wasserbeaufschlagung beobachtet worden. Hierbei kam es zu einer deutlichen Reduzierung der Transportraten sowohl für Wasser als auch für Gase. Eine speziell entwickelte Permeabilitätsmesskammer zur Durchführung von In- Situ-Messungen ermöglicht Untersuchungen zum Stofftransport unter axialer Zugbelastung bei verschiedenen Dehnungszuständen.

Large-scale virtualized data centers require cloud providers to implement scalable, autonomic, and energy-efficient cloud management systems. To address these challenges this thesis provides four main contributions. The first one proposes Snooze, a novel Infrastructure-as-a-Service (IaaS) cloud management system, which is designed to scale across many thousands of servers and virtual machines (VMs) while being easy to configure, highly available, and energy efficient. For scalability, Snooze performs distributed VM management based on a hierarchical architecture. To support ease of configuration and high availability Snooze implements self-configuring and self-healing features. Finally, for energy efficiency, Snooze integrates a holistic energy management approach via VM resource (i.e. CPU, memory, network) utilization monitoring, underload/overload detection and mitigation, VM consolidation (by implementing a modified version of the Sercon algorithm), and power management to transition idle servers into a power saving mode. A highly modular Snooze prototype was developed and extensively evaluated on the Grid'5000 testbed using realistic applications. Results show that: (i) distributed VM management does not impact submission time; (ii) fault tolerance mechanisms do not impact application performance and (iii) the system scales well with an increasing number of resources thus making it suitable for managing large-scale data centers. We also show that the system is able to dynamically scale the data center energy consumption with its utilization thus allowing it to conserve substantial power amounts with only limited impact on application performance. Snooze is an open-source software under the GPLv2 license. The second contribution is a novel VM placement algorithm based on the Ant Colony Optimization (ACO) meta-heuristic. ACO is interesting for VM placement due to its polynomial worst-case time complexity, close to optimal solutions and ease of parallelization. Simulation results show that while the scalability of the current algorithm implementation is limited to a smaller number of servers and VMs, the algorithm outperforms the evaluated First-Fit Decreasing greedy approach in terms of the number of required servers and computes close to optimal solutions. In order to enable scalable VM consolidation, this thesis makes two further contributions: (i) an ACO-based consolidation algorithm; (ii) a fully decentralized consolidation system based on an unstructured peer-to-peer network. The key idea is to apply consolidation only in small, randomly formed neighbourhoods of servers. We evaluated our approach by emulation on the Grid'5000 testbed using two state-of-the-art consolidation algorithms (i.e. Sercon and V-MAN) and our ACO-based consolidation algorithm. Results show our system to be scalable as well as to achieve a data center utilization close to the one obtained by executing a centralized consolidation algorithm.

A propriedade de auto-cura, em redes inteligente de distribuição de energia elétrica, consiste em encontrar uma proposta de reconfiguração do sistema de distribuição com o objetivo de recuperar parcial ou totalmente o fornecimento de energia aos clientes da rede, na ocorrência de uma falha na rede que comprometa o fornecimento. A busca por uma solução satisfatória é um problema combinacional cuja complexidade está ligada ao tamanho da rede. Um método de busca exaustiva se torna um processo muito demorado e muitas vezes computacionalmente inviável. Para superar essa dificuldade, pode-se basear nas técnicas de geração de árvores de extensão mínima do grafo, representando a rede de distribuição. Porém, a maioria dos estudos encontrados nesta área são implementações centralizadas, onde proposta de reconfiguração é obtida por um sistema de supervisão central. Nesta dissertação, propõe-se uma implementação distribuída, onde cada chave da rede colabora na elaboração da proposta de reconfiguração. A solução descentralizada busca uma redução no tempo de reconfiguração da rede em caso de falhas simples ou múltiplas, aumentando assim a inteligência da rede. Para isso, o algoritmo distribuído GHS é utilizado como base na elaboração de uma solução de auto-cura a ser embarcada nos elementos processadores que compõem as chaves de comutação das linhas da rede inteligente de distribuição. A solução proposta é implementada utilizando robôs como unidades de processamento que se comunicam via uma mesma rede, constituindo assim um ambiente de processamento distribuído. Os diferentes estudos de casos testados mostram que, para redes inteligentes de distribuição compostas por um único alimentador, a solução proposta obteve sucesso na reconfiguração da rede, indiferentemente do número de falhas simultâneas. Na implementação proposta, o tempo de reconfiguração da rede não depende do número de linhas nela incluídas. A implementação apresentou resultados de custo de comunicação e tempo dentro dos limites teóricos estabelecidos pelo algoritmo GHS. / The characteristic of self-healing, in smart grids, consists of finding a proposal for a reconfiguration of distribution system aiming at restoring the power, partially or completely to supply the network clients, in the event of network failure, which compromises the energy supply. The search for a satisfactory solution is a combinatorial problem whose complexity is proportional to the network size. An exhaustive search-based method is a time-consuming process and often computationally not viable. To overcome this difficulty, techniques for generating minimal spanning trees of the graph, which represents the smart grid, are exploited. However, the majority of studies in this area provide centralized implementations, where the solution for reconfiguration is achieved by a central control system. In this dissertation, we propose a distributed implementation, where each of the network switch collaborates in the development of the solution for reconfiguration. The proposed decentralized solution seeks a reduction in terms of the network reconfiguration time, in case of a single or multiple failures, thus increasing network intelligence. In this purpose, the GHS distributed algorithm is used as a basis for developing a self-healing solution to be embedded in the processing elements that are included within the line commutation switches of smart grid. The proposed solution is implemented using robots as processing units, which communicate via the same network, thereby creating a distributed processing environment. The several tested case studies show that, for smart grids that to have a single distribution feeder, the proposed solution allowed for a successful reconfiguration of the network, regardless of the number of simultaneous failures. In the proposed implementation, the network reconfiguration time does not depend on the number of buses and lines included. The implementation presents results of communication cost and time within the theoretical bounds of the GHS algorithm.

A propriedade de auto-cura, em redes inteligente de distribuição de energia elétrica, consiste em encontrar uma proposta de reconfiguração do sistema de distribuição com o objetivo de recuperar parcial ou totalmente o fornecimento de energia aos clientes da rede, na ocorrência de uma falha na rede que comprometa o fornecimento. A busca por uma solução satisfatória é um problema combinacional cuja complexidade está ligada ao tamanho da rede. Um método de busca exaustiva se torna um processo muito demorado e muitas vezes computacionalmente inviável. Para superar essa dificuldade, pode-se basear nas técnicas de geração de árvores de extensão mínima do grafo, representando a rede de distribuição. Porém, a maioria dos estudos encontrados nesta área são implementações centralizadas, onde proposta de reconfiguração é obtida por um sistema de supervisão central. Nesta dissertação, propõe-se uma implementação distribuída, onde cada chave da rede colabora na elaboração da proposta de reconfiguração. A solução descentralizada busca uma redução no tempo de reconfiguração da rede em caso de falhas simples ou múltiplas, aumentando assim a inteligência da rede. Para isso, o algoritmo distribuído GHS é utilizado como base na elaboração de uma solução de auto-cura a ser embarcada nos elementos processadores que compõem as chaves de comutação das linhas da rede inteligente de distribuição. A solução proposta é implementada utilizando robôs como unidades de processamento que se comunicam via uma mesma rede, constituindo assim um ambiente de processamento distribuído. Os diferentes estudos de casos testados mostram que, para redes inteligentes de distribuição compostas por um único alimentador, a solução proposta obteve sucesso na reconfiguração da rede, indiferentemente do número de falhas simultâneas. Na implementação proposta, o tempo de reconfiguração da rede não depende do número de linhas nela incluídas. A implementação apresentou resultados de custo de comunicação e tempo dentro dos limites teóricos estabelecidos pelo algoritmo GHS. / The characteristic of self-healing, in smart grids, consists of finding a proposal for a reconfiguration of distribution system aiming at restoring the power, partially or completely to supply the network clients, in the event of network failure, which compromises the energy supply. The search for a satisfactory solution is a combinatorial problem whose complexity is proportional to the network size. An exhaustive search-based method is a time-consuming process and often computationally not viable. To overcome this difficulty, techniques for generating minimal spanning trees of the graph, which represents the smart grid, are exploited. However, the majority of studies in this area provide centralized implementations, where the solution for reconfiguration is achieved by a central control system. In this dissertation, we propose a distributed implementation, where each of the network switch collaborates in the development of the solution for reconfiguration. The proposed decentralized solution seeks a reduction in terms of the network reconfiguration time, in case of a single or multiple failures, thus increasing network intelligence. In this purpose, the GHS distributed algorithm is used as a basis for developing a self-healing solution to be embedded in the processing elements that are included within the line commutation switches of smart grid. The proposed solution is implemented using robots as processing units, which communicate via the same network, thereby creating a distributed processing environment. The several tested case studies show that, for smart grids that to have a single distribution feeder, the proposed solution allowed for a successful reconfiguration of the network, regardless of the number of simultaneous failures. In the proposed implementation, the network reconfiguration time does not depend on the number of buses and lines included. The implementation presents results of communication cost and time within the theoretical bounds of the GHS algorithm.

La rupture différée des fibres SiC de type Hi-Nicalon à l’échelle multifilamentaire, des minicomposites de type Hi-Nicalon/PyC/SiC et Hi-Nicalon/type PyC/B4C a été étudiée à l’aide de moyens d’essais spécifiques et innovants. Des essais de fatigue statique sous air aux très hautes températures (900°C-1300°C) avec mesure des déformations ont ainsi pu être réalisés sur ces différents matériaux. Les résultats expérimentaux obtenus (durée de vie, déformation, lois de comportement en traction) ont permis de comprendre et de modéliser les mécanismes responsables de la rupture différée aux différentes échelles : - Les fils Hi-Nicalon rompent par mécanisme de fissuration lente activé par l’oxydation du carbone libre des fibres. Le mécanisme de fissuration est perturbé par la formation rapide d’oxyde SiO2 à partir de 1000°C : pour les faibles contraintes, la cinétique de fissuration lente est ralentie par formation d’oxyde protecteur empêchant l’accès de l’oxygène aux fissures ; pour les fortes contraintes, la rupture des fils est prématurée à cause de collages inter-fibres (fibre-oxyde-fibre). A 1200°C, le mécanisme de fluage semble être à l’origine de la rupture quasi-instantanée du matériau pour des contraintes supérieures à 200 MPa. - Les minicomposites Hi-Nicalon/type PyC/SiC rompent par mécanisme de fissuration lente ralenti par la présence de matrice SiC et par la formation d’oxyde SiO2 limitant l’accès de l’oxygène aux fibres. le mécanisme de fluage est observé à partir de 1200°C mais il n’a jamais été responsable de la rupture du matériau. - Les minicomposites Hi-Nicalon/type PyC/B4C rompent par mécanisme de fissuration lente ralenti par formation d’oxyde B2O3 à 900°C pour les fortes contraintes. Pour les autres températures et pour les faibles contraintes à 900°C le mécanisme de rupture est la diminution rapide du diamètre des fibres à cause de l’augmentation de la cinétique d’oxydation des fibres par l’oxyde B2O3. Des modèles analytiques basés sur ces différents mécanismes permettent de prévoir la durée de vie du matériau en prenant en compte les incertitudes de mesure et la variabilité des résultats de durée de vie. / Delayed failure of SiC Hi-Nicalon multifilament tows (500 fibers), minicomposites Hi-Nicalon/type PyC/SiC and Hi-Nicalon/type PyC/B4C was investigated in static fatigue, in air, at high temperatures (900°C – 1300°C) using specific and innovative devices. Static fatigue tests with measure of strain were performed on these materials. The experimental results (lifetime, strain, tensile behavior) have helped to understand and model the mechanisms responsible for the delayed failure at the different scales: - Hi-Nicalon tows rupture is caused by subcritical crack growth mechanism activated by oxidation of free carbon in the fibres. This phenomenon is disrupted by fast oxide SiO2 formation over 900°C: subcritical crack growth kinetic slows down for low stresses because of protective oxide formation which prevents the cracks from oxygen; For high stresses, the lifetime of Hi-Nicalon tows is weaker because of fibers interactions (fiber-oxide-fiber). At last, creep seems to cause the rupture of the tows for stresses over 200 MPa at 1200°C. - Hi-Nicalon/type PyC/SiC minicomposites break by subcritical crack growth slowed down by the SiC matrix and by the SiO2 formation which limit the access of the oxygen to the fibers. Creep occurs at 1200°C but it isn’t responsible of the rupture. - Hi-Nicalon/type PyC/B4C minicomposites break by subcritical crack growth slowed down by the formation of B2O3 oxide at 900°C for high stresses. The rupture is caused by the fast decrease of the diameter of the fibers at the other temperatures and for low stresses at 900°C. The oxidation kinetic of the fibers increases because of the dissolution of silica coating by B2O3 oxide. Analytical modeling was performed to schedule the lifetime of these materials and the variability of the experimental results is studied.