Visual field testing, or perimetry, measures peripheral visual loss in eye diseases such as glaucoma. Rarebit Perimetry (RBP) is a new and novel perimetric method, introduced in 2002 by Lars Frisén (2002), with the aim of detecting low degrees of neural damage within the retina. / RBP is unlike conventional perimetric methods that measure levels of retinal sensitivity, but instead uses very bright (i.e. suprathreshold) and very small targets to detect tiny areas of absolute blindness within otherwise normal areas of vision. RBP thus claims to locate miniscule gaps in the receptive field matrix of neurons in the retina, with the assumption that dead neurons leave gaps in this matrix. The most useful application of this idea is to detect progressive eye disease in the earliest stages (Frisén, 2002). Current research shows that RBP correlates with other standard visual field tests (Brusini, Salvetat, et al., 2005; Frisén, 2003; Gedik, Akman, et al., 2007; Martin & Wanger, 2004), but may afford greater sensitivity by detecting very mild visual losses missed by other tests (Martin, Ley, et al., 2004; Martin & Nilsson, 2007; Nilsson, Wendt, et al., 2007). / To date, there are no studies that definitively test the theoretical basis of RBP, so in this thesis I aim to explore the proposed underlying mechanisms and assumptions of this test. In particular, the proposed mechanism of RBP leads to specific predictions as to how responses will alter when the luminances of the RBP targets are systematically decreased. I therefore compared RBP responses of mean hit rate as a function of target luminance and found results to be inconsistent with the proposed RBP mechanism. Mathematical simulations were performed to explore reasons for the differences between the two groups (Chapter Seven).

L’objectif ultime de ce travail de thèse est d’évaluer la possibilité de valider et d’optimiser un processus de fabrication en utilisant la simulation numérique en tenant compte des incertitudes irréductibles sur le procédé, les matériaux et la géométrie du produit fabriqué. La prise en compte des incertitudes nécessite de quantifier les effets des variations des paramètres du modèle sur les sorties de celui-ci, en propageant ces variations via la simulation numérique pour évaluer leurs effets sur les sorties. Dans ce travail nous avons proposé une procédure pour déterminer le seuil de sensibilité du modèle numérique afin de construire des plans d’expériences numériques cohérents avec ce seuil. Nous avons également montré que, compte tenu des incertitudes sur les matériaux et la géométrie du produit, il est possible d’optimiser certains paramètres du procédé pour contrôler les effets des incertitudes sur les variations dimensionnelles et morphologiques du produit. Pour cela, nous avons proposé une procédure d’optimisation basée sur un algorithme NSGA-II et une méta-modélisation du procédé. L’application à l’emboutissage d’une tôle en U, retour élastique inclus, montre qu’il s’agit d’un problème de conception robuste pour lequel nous obtenons l’ensemble des compromis entre l’écart à la moyenne et l’écart type d’une fonction « performance » du procédé correctement choisie. Finalement l’analyse de ces résultats nous permet de quantifier le lien entre la notion de robustesse d’une solution optimisée du procédé et les critères de mesure de la qualité du produit / The ultimate objective of this thesis is to evaluate the possibility to validate and optimize a manufacturing process using numerical simulation and taking into account the irreducible uncertainties in the process, materials and geometry of manufactured product. Taking into account the uncertainties requires quantifying the effects of variations of model parameters on the outputs, by propagating these variations via computer simulation to assess their effects on the outputs. In this work, we have proposed a procedure to determine the sensitivity threshold of the numerical model to build numerical Design of Experiments consistent with this threshold. We have also shown that, given the uncertainties in the materials and the geometry of the product, it is possible to optimize certain process parameters to control the effects of uncertainties on the dimensional and morphological variations of the product. For this, we have proposed an optimization procedure based on NSGA-II algorithm and a meta-modeling of the process. The application for deep drawing of a U-shaped sheet metal part, springback included shows that it is a robust design problem for which we get all the compromise between the deviation from the mean and standard deviation of a "performance" depending on the process correctly chosen. Finally, the analysis of these results allows us to quantify the relationship between the notion of robustness of an optimized solution of the process and criteria for measuring the quality of the product