Reliability of electrical components is an issue studied to improve the quality of products, and to plan maintenance in case of failure. Reliability is measured by studying the causes of failure and the mean time to failure. One of the methods applied in this field is the study of component aging, because failure often occurs after degradation.
The objective of this thesis is to model the degradation of components in electrical engineering, in order to estimate their lifetime. More specifically, this thesis will study large area organic white light sources (OLEDs). These sources offer several advantages in the world of lighting thanks to their thinness, their low energy consumption and their ability to adapt to a wide range of applications. The second components studied are electrical insulators applied to pairs of twisted copper wires, which are commonly used in low voltage electrical machines.
First, the degradation and failure mechanisms of the various electrical components, including OLEDs and insulators, are studied. This is done to identify the operational stresses for including them in the aging model.
After identifying the main causes of aging, general physical models are studied to quantify the effects of operational stresses. Empirical models are also presented when the physics of degradation is unknown or difficult to model.
Next, methods for estimating the parameters of these models are presented, such as multilinear and nonlinear regression, as well as stochastic methods. Other methods based on artificial intelli­gence and online diagnosis are also presented, but they will not be studied in this thesis.
These methods are applied to degradation data of organic LEDs and twisted pair insulators. For this purpose, accelerated and multifactor aging test benches are designed based on factorial experimental designs and response surface methods, in order to optimize the cost of the experiments. Then, a measurement protocol is described, in order to optimize the inspection time and to collect periodic data.
Finally, estimation methods tackle unconstrained deterministic degradation models based on the measured data. The best empirical model of the degradation trajectory is then chosen based on model selection criteria.
In a second step, the parameters of the degradation trajectories are modeled based on operational constraints. The parameters of the aging factors and their interactions are estimated by multilinear regression and according to different learning sets. The significance of the parameters is evaluated by statistical methods if possible. Finally, the lifetime of the experiments in the validation sets is predicted based on the parameters estimated by the different learning sets. The training set with the best lifetime prediction rate is considered the best.