Dans de nombreux domaines, tels que l’industrie des transports ou bien des infrastructures, la tendance est à l’introduction toujours plus importante d’équipements électriques. De ce fait, les industriels sont de plus en plus confrontés à la nécessité de fournir des dispositifs robustes et fiables avec un minimum de maintenance. Les composants électroniques, tels que les transistors IGBTs ou MOSFETs et les diodes rassemblés dans des modules de puissance, sont au cœur de la conversion d'énergie électrique. En conséquence, ils sont soumis en opération à de fortes contraintes environnementales et fonctionnelles (température, humidité…). L’ensemble de ces contraintes a un impact sur la durée de vie des composants, et donc sur la fiabilité des dispositifs. D’un point de vue économique, le remplacement d’un équipement défectueux est moins pénalisant qu’une défaillance brutale du système. Ainsi, l’utilisation d’outils de diagnostic est nécessaire pour prédire la durée de vie restante des dispositifs en opération, et mettre en place une maintenance adaptée et efficace.Pour déterminer la durée de vie restante des modules de puissance en opération, des modèles de durée de vie sont utilisés. La plupart de ces modèles sont établis soit de manière empirique, soit de manière physique, soit de manière statistique. Les modèles empiriques sont les plus courants, car leur réalisation et implémentation sont maintenant bien connues. Ils se basent sur des résultats issus de tests de cyclage accélérés qui reproduisent les contraintes endurées par le module de puissance sous des conditions "accélérées" de fonctionnement. Une extrapolation est ensuite nécessaire pour obtenir l’état de santé du dispositif dans des conditions normales de fonctionnement. Le principal inconvénient de ces modèles réside dans le manque de description des mécanismes physiques responsables de l’endommagement. Ce manque peut mener potentiellement à des erreurs, notamment lors de l’extrapolation. C’est pourquoi les modèles basés sur la physique connaissent un intérêt grandissant.Dans cette thèse, deux modèles de durée de vie basés sur la physique et appliqués aux modules de puissance IGBTs sont proposés et comparés. La première approche est basée sur les déformations induites à l’intérieur de l’assemblage du module lorsque soumis à des contraintes thermiques. Dans ce cas, la dégradation est décrite via la quantification des déformations pour un stress thermique donné. Dans la seconde approche, le modèle de durée de vie est basé directement sur l’endommagement via l’établissement d’un modèle de dégradation. La comparaison des deux modèles met en lumière les défauts et qualités de chacun. D’une manière plus générale, l’établissement et la comparaison de ces modèles s’inscrit dans une démarche de développement d’outils de diagnostic afin de prédire la durée de vie restante des modules de puissance en opération. / The domain of power electronics reliability has become an important center of interest with the recent massive system electrification. The manufacturers are more and more confronted to the necessity of producing reliable devices with optimized maintenance. Electronics components, such as IGBTs, diodes and MOSFETs assembled in power modules, are at the center of the systems conversion, and as a consequence, are subjected to high environmental and functional stresses (ambient temperature, vibrations…). All these factors have a strong impact on the components lifetime and thus on the devices reliability. Economically, scheduling a maintenance with a system replacement is less detrimental than a brutal failure of the system. As a consequence, the use of lifetime prognostic tools is necessary. The problematic consists in the health state prediction of power modules in functioning to be able to schedule a maintenance before the failure of the equipment.To be able to determine the remaining useful lifetime of power modules in functioning, lifetime models are used. These models can either be empirical, physical or statistical. The empirical models are the most common ones, because of their easy establishment and implementation. They are based on results from accelerated power cycling tests, which reproduce the stresses endured by the power modules in severe conditions. An extrapolation is then needed to obtain the power module health state in normal functioning conditions. The main drawback of these models is the lack of description of the physical mechanisms leading to damage, resulting potentially in errors in particular during extrapolation. That’s the reason why physical models start to draw more attention.In the thesis, two physical lifetime models of IGBT power modules are proposed. The first approach is based on deformation induced inside the device assembly in operation. The degradation is in this case described by the quantification of deformation related to thermal stresses. In the second approach, the lifetime model is based directly on damage through the establishment of a degradation model. These two lifetime models are finally compared to show the benefits and disadvantages of each. More generally, the establishment and comparison of these models is part of an approach to develop diagnostic tools so that the remaining useful lifetime of power modules can be predicted in operation.

Gesintertes Silber (SAG) stellt eines der vielversprechendsten Materialien für Hochtemperaturanwendungen in der Leistungselektronik dar. Im Vergleich zu konventionellen Loten sind die mechanischen und thermischen Vorteile enorm, allerdings hochgradig prozessabhängig. Zusammen mit den relativ zeitintensiven Ermüdungstestmethoden ist das die Ursache, dass es aktuell nur wenige Lebensdauermodelle dazu gibt. In dieser Arbeit wird am Beispiel solcher SAG-Proben ein mechanisch beschleunigter, isothermer Biegeversuch vorgestellt, welcher das Potenzial hat, die zeitkritischen Temperatur- oder Lastwechselversuche zu ersetzen. Zum Vergleich wurde ein Temperaturwechseltest als Referenzversuch durchgeführt. Hierzu wird zunächst der Stand der Technik des Silber-Sinterns aufgezeigt, wobei der Schwerpunkt auf der mechanischen Materialcharakterisierung liegt.Wo das elastische Verhalten als näherungsweise allein porositätsabhängig gelten kann, ist die inelastische Dehnung noch unzureichend untersucht. Besonders die zeitabhängige inelastische Dehnung (Kriechen) zeigt noch kein vollständig konsistentes Bild, wodurch auch die Fehlermechanismen und deren Gewichtung noch nicht grundsätzlich als geklärt gelten können. Die gängigsten Belastungstests, welche in der Literatur zu finden sind, haben schwerwiegende Nachteile. Der hohe Zeitbedarf, die teils schwer quantifizierbaren Fehlerparameter und die fehlende Einstellmöglichkeit des Verhältnisses Kriechdehnung zu plastischer Dehnung sind hier im Besonderen zu nennen. Der rein dehnungsgesteuerte Biegeversuch hat diese Nachteile nicht. Über die Biegegeschwindigkeit ließe sich der Kriechanteil nahezu beliebig erhöhen (ggf. unter Nutzung von Haltezeiten). Die Biegeversuche wurden isotherm bei fünf Temperaturen von 22◦C bis 125◦C mit je drei Amplituden und drei Biegegeschwindigkeiten durchgeführt. Schlecht gesinterte Proben machten sich reproduzierbar als Frühausfall bemerkbar, so dass sich die Methode bereits gleich zu Beginn als hervorragender Qualitätstest bewährte. In puncto Ermüdung konnte ein stabiles und reproduzierbares Ausfallverhalten in Form von vergleichbaren Weibull-Formfaktoren und Ausfallbildern in den metallografischen Schliffen gefunden werden. Mit den Daten der Biegeversuche wurde ein fehlerphysikalisches Lebensdauermodell (Coffin-Manson) aufgestellt, welches erfolgreich den Ausfall des Temperaturwechseltests als Referenzversuch vorhersagen konnte. / Sintered silver (SAG) as die attach material is one of the promising solutions to exploit the advantages of high-gap semiconductors in power electronics. The mechanical and thermal properties are far superior to solders, but severely process-dependent. Combined with the time requirements of the state of the art (SoA) fatigue test methods this is most likely the reason for the lack of profound reliability studies yet. This thesis presents an isothermal bending test, which has the ability to replace the time-consuming thermal shock test as primary fatigue experiment for physics of failure based (PoF) lifetime models. A benchmark against a conventional thermal cycling test was done. The state of the art of the silver-sintering technique will be given with focus on the mechanical material characterization. While the elastic properties are mostly porosity-dependent, the inelastic properties are insufficiently examined yet. Especially the creep does not show a consistent image, what leads to many questions regarding the failure mechanism. The most common fatigue tests in the literature do have serious disadvantages. The time-consumption is high, the failure parameter can hardly be quantified and the ratio of plasticity to creep cannot be adjusted easily. The pure mechanical bending test does not have those disadvantages. By changing the bending-speed and the addition of holding times, the creep can be adjusted almost at will. The bending-experiments were conducted at five different temperatures between 22°C and 125°C and with three bending amplitudes as well as three speeds. Insufficiently sintered samples could be identified very early. This already proofed the value of the test as a quality test. Furthermore, a stable and repeatable fatigue behaviour could be observed, what was given by stable Weibull-exponents and repeatable cross sections. A lifetime-model was established by usage of the bending-test-data, what eventually predicted successfully the lifetime of a thermal cycling reference test.

Consider life testing of J-component series systems that are subjected to stress levels that are steeper than that at normal usage condition. The objective of performing such life tests, commonly known as Accelerated Life Testing (ALT) in the literature, is to collect observations on system failure times within a limited time frame. The accelerated observations are then used to infer on the component and system reliability metrics at usage stress. In this thesis, the existing literature is first extended by considering the general case of K stress variables, as opposed to the usual consideration of a single stress variable. Next, a general model assuming that the component log-lifetimes belong to an arbitrary location-scale family of distributions, is formulated. The location parameters are assumed to depend on the stress variables through a general stress translation function, while the scale parameters are assumed to be independent of the stress variables. This formulation covers the standard lifetime distributions as well as well-known stress translation functions as special cases. Bayesian methodologies are then developed for four special cases of the proposed general model, viz., exponentials, Weibulls with equal shape parameter, Weibulls with distinct shape parameters and log-normals with distinct scale parameters. For exponential and Weibull models, the priors on lifetime parameters are assumed to be log-concave and independent of each other. The resulting univariate conditional posterior of each lifetime parameter given the rest, is shown to be log-concave. This facilitates Gibbs sampling from the joint posterior of lifetime parameters. Propriety of the joint posteriors with Laplacian uniform priors on stress coefficients are also proved under a suitable set of sufficient conditions. For the log-normal model, the observed data is first augmented with log-lifetimes of un-failed components to form complete data. A Gibbs sampling scheme is then developed to generate observations from the joint posterior of lifetime parameters, through the augmented data and a conjugate prior for the complete data. In all four cases, Bayesian predictive inference techniques are used to study component and system reliability metrics at usage stress. Though this thesis mainly deals with Bayesian inference of accelerated data of series systems, maximum likelihood analysis for the log-normal component lifetimes is also performed via an expectation-maximization (EM) algorithm and bootstrap, which are not available in the literature. The last part of this thesis deals with construction of optimal Bayesian designs for accelerated life tests of J-component series systems under Type-I censoring scheme. Optimal ALT plans for a single stress variable are obtained using two different Bayesian D-optimality criteria for exponentially distributed component lives. A detailed sensitivity analysis is carried out to investigate the effect of different planning inputs on the optimal designs as well.